The AI Optimist

I remember it so clearly, the Exhibitor’s booth with the simple name Ruth at TechCrunch Disrupt 2024, away from the posturing about AGI and the next trillion-dollar unicorn.

Megs Shah and Sandy Skelaney are doing something different and non-profit – Ruth is an AI app that does not track its users. And helps them use technology to protect themselves.

Ask Ruth - Your Digital Guardian - Here

Delivers trusted information to people confused about harassment and bullying, in danger from abuse and trafficking, and technically exposed in ways they don’t know.

Their creation is RUTH, a digital guardian helping people navigate the dark corners of tech-facilitated abuse.

The story of how RUTH came to be reveals the human side of artificial intelligence, how sometimes the most powerful solutions come from personal experience.

"I was trying to keep up with my son playing video games.

I couldn't figure out what he was doing on there, and I was worried about sextortion.

So I went on to OpenAI and asked how to talk about sextortion with your teenager."

The response? "Sorry, it's out of policy."

For Megs, a seasoned technologist, this wasn't just an inconvenience – it was a call to action.

"The technologist in me didn't accept that answer," she says with a slight smile.

"I was like, oh hell no, this can't be."

That moment of frustration in 2020 led to what would become Parasol Cooperative and RUTH.

The drive to move from AI concept to reality requires more than technical expertise. It demands a deep understanding of the human cost of technology abuse.

Sandy Skelaney, Parasol's COO, frames the challenge:

"The field of tech-enabled abuse has grown exponentially.

We're talking about sextortion, deepfakes, revenge porn.

Organizations working with survivors of domestic violence report that 98% of the survivors they're working with are experiencing technology-related abuse."

What makes RUTH different isn't just its focus on tech abuse – it's how it approaches the problem.

Most AI applications harvest user data, RUTH deliberately maintains zero knowledge of its users.

When you exit the chat, your data vanishes. No emails collected, no names stored, no digital breadcrumbs left behind.

"I've been in data for God knows how long," Megs explains,

"and I've been on the side of

'let's collect every possible piece of information we can about a person.'

But being on the survivor side of the mindset now, I've realized that could revictimize people.

The goal for us is ensuring we're not participating in that."

This privacy-first approach isn't just an ethical choice – it's a safety feature.

For people facing tech-enabled abuse, being tracked by another app, even one trying to help, could put them at risk. RUTH's design ensures that they can find help without leaving a digital trail.

Creating an AI that's both helpful and trauma-informed required rethinking how AI should interact with new users.

"Ruth is very clear that they're providing sensitive responses, making sure they're checking in with your safety," Sandy explains.

"But also at the same time,

'I'm still an AI, I'm not a human being.'"

The technical implementation of RUTH reflects this careful, human-centered approach.

While many AI companies rush to fine-tune their models with user data, Parasol made the conscious decision to build differently.

They chose Anthropic's model as their base after extensive testing, appreciating its robust responses and customization potential.

"We could have easily just put a chatbot out there and said, 'hey, try it out,'" Megs explains.

"But we are very much survivor focused. In everything we do, we try to consider:

how would this make me feel if I was the one asking these questions?"

This led to months of careful development between their soft launch in June and their wider release in September 2024.

The team worked on prompt engineering, testing, and validation to ensure RUTH's responses felt more like talking to a supportive friend than a clinical professional.

They focused on breaking down language barriers, essential in making support available to immigrant communities.

RUTH's trauma-informed approach manifests itself in subtle but crucial ways.

Before diving into solutions, it checks if users are in a safe place and whether their devices might be compromised.

It acknowledges the gravity of situations without inducing fear, offering practical guidance while ensuring users maintain agency over their choices.

The privacy architecture is equally thoughtful.

"When you go to our site to use Ruth," Megs emphasizes,

"we do not collect any emails, we do not collect your name.

We even tell you don't give us all this information. We don't need it."

This zero-knowledge approach extends to their development process – they don't even use chat data to improve the model, a stark contrast to standard AI development practices.

Instead of relying on user data for improvements, RUTH uses a sophisticated RAG (Retrieval Augmented Generation) architecture that pulls from carefully vetted sources.

This allows them to update their knowledge base with the latest research and safety information without involving user privacy.

They've built their backend to be modular, allowing them to switch out AI models as better options emerge, while maintaining their strict privacy standards.

The global expansion of RUTH happened almost by accident, but its rapid adoption speaks to a universal need.

From its initial testing phase, the team discovered RUTH was providing accurate resources and support not just in major cities, but in remote locations across the globe.

"We've tested all over the world," Sandy notes,

"in Kandahar, Afghanistan, Algiers – and it seems to be

giving accurate resources and information."

This international reach reveals how tech-facilitated abuse transcends borders and cultures.

In each new country, RUTH adapts its responses to local context while maintaining its core principles of privacy and trauma-informed support.

The team has been particularly moved by how different communities have embraced the tool in ways they never anticipated.

For victim service professionals worldwide, RUTH has become an invaluable ally.

"Advocates are trained in mental health," Sandy explains, "they are not trained in digital safety planning. Being able to give them a tool so that they can sit down and do a case plan with the survivors they're working with... it absolutely supercharges that work."

The feedback loop is designed to maintain user privacy while allowing for improvements. Users submit anonymous feedback about incorrect or unhelpful responses, which the team prioritizes for updates.

This allows them to refine RUTH's responses without compromising their zero-tracking commitment.

"Every day technology is changing and evolving," Sandy notes.

"Every day there's new ways people are finding to abuse.

How do you train it to be trauma informed?

How do you give it the right information in terms

of the intersection of tech and abuse?"

Their solution was to create a system that could be updated rapidly with new information while maintaining the emotional intelligence and safety-first approach that makes RUTH unique.

In less than a year, what started as a tool to help parents discuss online safety with their children has evolved into a comprehensive resource being used by individuals, families, and professionals across 29 countries.

It provides support in situations ranging from basic digital safety questions to complex cases of tech-facilitated abuse.

In a tech landscape focusing on data collection and monetization, Parasol Cooperative is showing that it's possible to build powerful AI tools prioritizing privacy and safety above all else.

RUTH represents a different vision of what AI can be.

Today AI discussions often center on replacing humans, RUTH aims to enhance human connection and support.

It's not trying to be human – it's trying to be transparently artificial.

"Knowledge is power that no one can take away from you," Megs says, summing up their mission.

"And if we're able to equip you with that, then we've done our job right, because we've trained you to be more vigilant about yourself and your family online."

For those interested in supporting this work, Parasol Cooperative accepts donations through their website – and as Megs emphasizes, every penny goes directly to making these tools freely accessible to those who need them.

In a conference full of startups promising to change the world, I found one that might – not through grand pronouncements about artificial general intelligence, but through the simple act of helping people feel safe in an increasingly complex digital world.

That simple act is likely worth more than a unicorn.

Especially to those, it helps.

Thanks for reading The AI Optimist! This post is public so feel free to share it.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.theaioptimist.com

In the midst of the AI glory days, in a room buzzing with the impact and momentum of the industry, no one seemed to care about copyrights except the speaker on the Disrupt stage.

While watching Martin Casado of A16Z deftly handle questions about AI regulation – he stopped me by noting the importance of copyright to regulation.

The room packed with the usual entrepreneurs, investors, and journalists eager to hear how Silicon Valley will lead the AI revolution. But something wasn't sitting right with me.

"Why do tech companies treat AI training like a copyright-free zone?"

I finally asked, cutting through the careful corporate speak.

The room went quiet. Casado's response was telling - he immediately reached for the comfortable analogy of Napster and how peer-to-peer sharing eventually led to Spotify and Apple Music.

As someone who lived through that transformation, I knew the punchline he was missing: musicians now earn a fraction of what they once did.

We're about to repeat history, but this time it's not just music - it's every form of human creativity.

Copyright is The Castle, AI has Wings

Copyright law is like a rundown medieval castle. For centuries, it protected creative works with its stone walls, drawbridges, and vigilant guards.

Then AI showed up - not with battering rams, but with wings. Our carefully constructed walls became meaningless overnight.

"The old rules weren't built for machines that can process centuries of human creativity before breakfast,"

"We're using rules written for printing presses to regulate digital shapeshifters."

What if one day a talented artist receives a cease-and-desist letter for AI-generated art that infringes on her own style.

It’s possible to be legally challenged over AI mimicking her creativity, while the same AI had likely trained on her work without permission or compensation.

The Breaking Point Is Here

The collision between old laws and new technology isn't theoretical - it's happening right now. At TechCrunch Disrupt, I watched this play out in real time.

When tech leaders discuss AI training, they treat copyrighted work like it's a free-for-all resource.

But for creators watching their life's work being ingested into AI models without compensation or consent, it's not so simple.

This is where we hit two colliding truths:

Engineers say: "Input isn't copyrightable"

Creators say: "That's my life's work"

Both are right. Both are wrong. And that's exactly why we need a new approach.

Beyond Protection: The Participation Economy

Copyright law was built for a world of physical delivery - books, records, paintings you could hold in your hands.

It was about protecting specific copies from unauthorized duplication.

But in today's digital age, where AI can process and transform millions of works instantly, that framework isn't just outdated - it's obsolete.

The solution isn't more lawsuits or stronger copyright walls. It's about building a new system that benefits both creators and AI development.

* LLMs need quality training data to build better language models.

* Creators need fair compensation and recognition for their work.

*  Instead of seeing these as competing interests, we can align them.

I've developed two potential frameworks to help creators and AI interact. Not through courts and litigation, but through code and collaboration.

These solutions focus on measuring and rewarding actual influence - how much a creator's work contributes to AI outputs and model improvement.

The Token Economy: Rethinking Creative Value

Thinking about the comparison between today's AI to the Napster era, it crystallized something for me.

We're not just facing a copyright problem - we're facing a value recognition problem.

But unlike the music industry's painful transition to streaming, we have a chance to build something better from the start.

The Token Economy model I'm proposing treats creative works like living assets rather than static files. Here's how it works:

Influence Tracking in Action

Imagine each piece of creative content broken down into tokens - not just words or images, but meaningful building blocks that carry their influence forward.

Every time an AI model learns from or uses these tokens, it creates a trackable imprint.

This isn't theoretical - we're already seeing similar principles at work in how transformer models process and weight different inputs.

Smart Compensation That Scales

Let's talk real numbers:

* Training an AI model costs roughly millions to hundreds of millions:

* Current licensing deals max out around $20-100 million total for larger businesses

* Most creators earn pennies per view/use of their content

Instead of flat licensing fees or per-use payments, make compensation dynamic:

* High-impact content (like authoritative research papers) earns premium compensation

* Generic content (like basic product descriptions) receives minimal compensation

* Unique creative works (like distinctive writing styles or innovative code) earn mid-tier compensation

The Efficiency Breakthrough

This model doesn't just serve creators - it makes AI development more efficient. Here's why:

* Better training data leads to better models

* Transparent compensation reduces legal risks and costs

* Standardized systems lower content acquisition expenses

What if compensation was based on actual influence and value created?

The metric would serve the business and a select group of creators would be rewarded financially, others with credits and bonuses.

Improving AI Training with Fair Participation

When we talk about LLMs and training data, quality matters more than quantity. Right now, AI companies are scraping everything they can find, treating all content as equal.

But we know that's not true. Some content - whether through its clarity, originality, or influence - contributes more to model performance than others.

Performance-Based Value Creation

Here's what happens when we shift from protection to participation:

* Value Tiers That Make Sense:

* Research papers that improve model accuracy

* Creative works that enhance output quality

* Technical content that improves specialized knowledge

* Cultural works that help models understand context

Each tier gets compensated based on measurable improvements in model performance. It's not just about paying for content - it's about investing in quality.

The Technical Framework

I've looked at a  proof-of-concept system that tracks how AI models pay attention when they create.

Think of it like a heart rate monitor for creativity. Every time the AI focuses on certain parts of its training, we can trace those patterns. This isn't perfect attribution, but it's measurable impact.

The code does something fascinating:

* Watches attention patterns across neural networks

* Identifies distinctive signatures in content influence

* Creates trackable metrics for compensation

Download the Code

Download the Claude discussion PDF

Real-World Applications

We're already seeing similar ideas emerge:

* Stability AI exploring creator compensation models

* Adobe's Content Credentials system tracking image origins

* Credtent empowering creators to control how LLMs use their work

Digital Influence Rights: Beyond Traditional Copyright

The challenge with traditional copyright isn't just technological - it's conceptual.

We're trying to apply rules meant for copying books to systems that transform and recombine information in entirely new ways.

Instead of fighting this transformation, we need to embrace and shape it.

What Are Digital Influence Rights?

Digital influence rights represent a fundamental shift in how we think about creative value in the AI era. Instead of focusing on preventing copying, these rights center on:

* Measuring how content shapes AI outputs

* Tracking the impact of creative work on model performance

* Creating value through influence rather than restriction

Core Mechanics

* Value-Based Tiers

* High-impact content earns premium rights

* Cultural significance affects earnings

* Dynamic pricing based on actual use and influence in improving AI outputs

* Creator Benefits Program

* AI system credits that grow with contribution

* Premium access rights to AI tools

* Direct influence over future AI training decisions

* Democratic Access

* Small creators can participate without legal overhead

* Impact-based earnings tracked through code

* Technical solutions replace legal battles

Technical Implementation

The system works by:

* Tracking attention patterns in neural networks

* Measuring how different inputs affect model performance

* Creating transparent metrics for value distribution

While the code for these systems is still emerging, we're seeing similar concepts being developed across the industry.

For those interested in learning more about these developing technologies, I recommend following:

·         Technical discussions about AI attribution systems

·         Open source AI model development communities

·         Creator rights organizations working on AI attribution

·         Companies developing content credentialing systems

Tracing AI's Creative DNA: Starting the Conversation

When I talk about tracking creative influence in AI, some say it's impossible - like trying to find a specific drop of water after it's been poured into the ocean.

They're not wrong about the complexity, but they're missing an important point.

We don't need perfect attribution to start building better systems.

Netflix doesn't know exactly why you watched that show, but they can make educated guesses about what influences your choices.

Spotify doesn't perfectly understand music but can track patterns of influence and similarity well enough to build a business model.

I'm not an engineer, but I've been exploring these concepts with AI tools to understand the possibilities.

The proof-of-concept ideas I'm working on are simple starting points - ways to begin thinking about how we might track and value creative influence in AI systems.

 It's like watching the attention patterns of AI models - where do they focus? What influences their outputs?

This isn't about building a perfect system. It's about starting a conversation and exploring practical first steps toward fair attribution and compensation.

The New Creator Economy: Participation Over Protection

We're not just facing a copyright crisis - we're witnessing the birth of a new creative economy.

One where the question isn't whether AI will use your work, but how you'll participate in shaping its development.

When an AI model ingests creative work, that influence disappears into a black box.

Creators lose connection to their work's impact, while AI companies gain value without sharing it. But it doesn't have to be this way.

Instead of focusing on restricting AI's use of creative work, we need to build systems that:

* Track real influence over AI outputs

* Reward meaningful contributions to model improvement

* Create new opportunities for creator participation

This isn't about stopping AI - it's about making it work better for everyone.

When creators see and benefit from their influence on AI systems, they're more likely to contribute quality work. When AI companies measure and reward valuable contributions, they build better models.

By measuring influence and creating clear value pathways, we can build a creative economy that works in the age of AI.

Building Tomorrow's Rules: Performance Over Protection

Traditional copyright law looks increasingly like a relic - rules written for printing presses trying to govern quantum computers.

The solution isn't more restrictive laws. It's smarter systems.

The core challenge is clear. Current copyright frameworks focus on controlling copies, but AI doesn't just copy - it learns, transforms, and generates.

We need rules built for this new reality. Rules that focus on:

* Measuring real impact rather than restricting access

* Rewarding performance instead of policing permission

* Creating value through participation, not protection

We're seeing the first signs of this transformation in how AI companies approach training data.

The question isn't whether they'll use creative work - it's how to make that use fair and beneficial for everyone involved.

The future won't be built in courtrooms. It will emerge from new systems that:

* Track and reward creative influence

* Enable small creators to participate without legal battles

* Allow innovation and fairness to coexist through smart technology

The question isn't if we'll move beyond traditional copyright - it's how quickly we'll build something better.

AI for Creators and Developers - A Model for Discussion

Opening New Possibilities: Beyond Copyright Litigation

"Amazing stuff doesn't come from the middle soup that generative AI will create."

This is crucial to understand - we're not just talking about compensation, we're talking about finding those edge cases that create extraordinary results.

Let me be clear about where we stand:

"We don't need better copyright laws.

We need a new system that turns creators from adversaries into stakeholders."

The question isn't whether AI will use your work - it's whether you'll help shape how this happens.

Here's what's possible right now:

For Creators:

"You won't immediately get paid for AI using your work, but you will have a way of understanding how it does."

This transparency is the first step toward fair participation.

For Engineers:

Yes, we recognize the limitations of attention-based attribution. But that's exactly why we need your help to:

 "make something better because we could do things here that could really help, not only help creators compensate them, but also find those edge cases that make the amazing stuff."

A System of Aligned Incentives:

Instead of random licensing fees, imagine a system where:

* "Usage influence tracking" quantifies how creator input affects specific outputs

* Higher influence leads to higher credits, creating relevance-based compensation

* "Think of this as intellectual currency" - creators don't lose control, they gain influence

"This isn't about lawsuits. It's a tech-enabled system where content usage aligns with incentives for creators."

Building Future to Cooperate instead of compete:

* Value-based tiers recognizing different levels of impact

* An AI-backed loyalty program offering both monetary and non-monetary rewards

* Leveling the playing field for smaller creators through collective impact

We're witnessing the birth of something new - a system where creators and AI can grow together rather than fight each other.

Instead of wasting resources on copyright litigation - which requires deep pockets and armies of lawyers - we can build technology that aligns incentives, tracks influence, and rewards meaningful contributions.

The future isn't about protecting content from AI; it's about participating in how AI evolves.

Whether you're a creator wondering about your AI influence or an engineer interested in these technical challenges, there's a role for you in shaping this future.

The tools are emerging. The possibilities are clear.

Now it's time to move beyond the copyright battles of yesterday and build something better together - a system that turns creators from adversaries into stakeholders in AI.

Thanks for reading The AI Optimist! This post is public so feel free to share it.

RESEARCH PHRASES

AI copyright, copyright law AI,AI and copyright, AI legal issues, AI generated content copyright, copyright protection for AI, legal implications of AI,AI ownership rights, AI and intellectual property, copyright law for AI generated art, can AI own copyright, legal framework for AI creativity, AI and fair use challenges of copyrighting AI output, AI copyright law trends, how AI is changing copyright law, AI copyright law cases, AI copyright law legislation, future of AI copyright law

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.theaioptimist.com

From Art to AI: How Generative Models Are Reinventing Advertising

The advertising industry is about to experience its AWS moment.

Just as cloud computing transformed IT from a capital-intensive hardware business into an API-driven software model, AI is pivoting advertising from a labor-intensive creative process into a model-driven generation engine.

At the center of this Generative Experience experiment stands Hikari Senju, a Harvard-educated technologist who saw what others missed: the future of advertising isn't about better targeting or fancier platforms—it's about teaching machines to understand and evolve brand DNA.

For the past 50 years, advertising has operated like a game of telephone played across conference rooms.

Creative teams pitch concepts to brand managers, who relay requirements to designers, who pass assets to media buyers, who then try to translate everything into campaigns.

Each handoff introduces friction, loses context, and dilutes the original creative vision. The result? Billions in wasted ad spend and brands that feel more like committees than personalities.

It's a system that somehow survived into the digital age, but like all paper-driven processes, it's about to be washed away by generative AI.

What Hikari and his team at Omneky have built is effectively a "generative brand management" platform—think GitHub for brand evolution.

Instead of managing static assets and style guides, they're building and refining brand models that can learn, adapt, and generate in real-time.

These models don't just create ads; they absorb performance data, understand consumer engagement, and continuously evolve the brand's voice across every platform.

It's the difference between having a rulebook and having a living, breathing brand personality that can hold millions of simultaneous conversations while staying true to its core identity.

This isn't just a better way to make ads—it's the foundation for how all brands will express themselves in the AI age.

The Artist and the Algorithm

Picture a young computer science student at Harvard, drawn to machine learning and art, who spots something others missed: the future of advertising isn't in better targeting or fancier platforms—it's in teaching AI to be creative.

This is the story of Hikari Senju and how his unique vision is transforming advertising from a human guessing game into a data-driven science.

"I grew up in Westchester, New York. My grandfather was an executive at IBM, my dad's an artist. From a very early age, I had this environment full of art and creativity as well as technology, computer robotics competitions as a kid." - Hikari Senju

While studying at Harvard from 2011 to 2014, Senju encountered early generative models that would change his perspective.

"It was there when I saw some of the early generative models, some of the precursors to generative adversarial networks. I got very excited about the space. Particularly because of the idea of an AI or technology being artistic and being creative, generating art. I found even the possibility of that incredibly exciting."

The seed was planted: what if AI could bridge the gap between creative expression and data-driven results?

The Birth of a Data-Driven Content Generation Vision (2018)

After an early success with an ad tech company in college, Senju noticed advertising's biggest problem wasn't distribution—it was developing the creative.

“More than half of marketing budgets are wasted every year on ineffective content.

The quality of the design, the quality of the creative is the biggest impact in terms of driving sales.

And yet, people are throwing darts in the dark."

This realization led to the founding of Omneky in 2018, with a laser focus on creative content generation. As Senju explains:

“First off, it was the biggest lever, the step change of this technology would enable.

And I think as a startup, you want to pick your bets carefully... If it's just incremental productivity boost, then it kind of favors incumbents who have those distribution advantages."

The Science Behind the Magic

What makes Omneky's approach unique isn't just generating content—it's building living, breathing brand models that learn and adapt.

The system works through a sophisticated process that Senju breaks down into several key components:

### 1. Brand DNA Extraction

“We train on the corpus of the customers' historical content that they've published, their social media, their brand books.

We have a whole brand management page for customers to upload all the brand assets and give it brand rules and brand guidelines."

### 2. Rule-Based Learning

The system learns not just what to do, but what not to do:

“What words can you not say? What words can you say when you're generating imagery?

What kinds of imagery are you seeking?

What are the rules when it comes to displaying your logo that you cannot violate?"

### 3. Dynamic Optimization

“Our edge is really not just about generating the creative content, but linking it to the data from the distribution platform.

What types of content is having higher click-through rates and return of ad spend based on the various ad networks and ad platforms, how do we incorporate that data into the generation process?"

### 4. Creative Brief Innovation

The platform changes how creative briefs are developed:

“We have a creative brief tool that generates ideas for consumers to test.

Customers will input the objectives of their campaigns, the audiences that they're targeting.

And then we will suggest the platforms to advertise.

We will suggest the copy, imagery. We'll even generate sample mood board imagery and sample product photographs."

Beyond Digital Asset Management

The traditional world of static brand guidelines is giving way to what Senju calls "generative experience management":

“In the old world that we used to live in, it was the world of digital asset management—static brands, static assets.

But we're going to live in a world in the future where all the assets are going to be generated on the fly in real-time, based on personal data, based on what's trending, based on all the new learnings."

This shift changes how brands maintain consistency while staying dynamic:

“Brands want to make sure that as they're scaling that brand experience, that they're also regaining control of that and that brand experience is consistent across different advertising platforms, across different social platforms, across different display platforms or streaming platforms."

The AI Agent Revolution

The latest is Omneky's move toward autonomous AI agents that can manage entire advertising campaigns:

“We're working on AI agents for automating media buying and helping businesses be even more agile when it comes to launching and optimizing creative content.

The AI agents can be more agile and run 24/7, to take into account real-time signals of the market."

Breaking Down Organizational Silos

Senju identifies a critical problem in traditional advertising structures:

“Think about the inefficiency in organizations today when it comes to creative, especially if you're a larger company. You tend to have a marketing department and a creative department.

Those two departments don't necessarily speak the same language.

The marketing team tends to be more involved in data.

The creative team tends to be more involved in stories of the big idea."

The solution? AI agents that bridge this gap:

“In the future, AI systems will be way more efficient. An AI system will in real time incorporate data from marketing into the generation process.

And you won't have to have cross-team communication and layers of people communicating with each other to execute an idea. It'll be in real time."

Democratizing Advertising

One of the most profound impacts of this technology is how it levels the playing field for startups:

“You can run an effective advertising campaign with something as small as a $100 budget, generate some effective creative and run that as an ad and get potentially dozens of customers.

The entry ticket for running an advertising campaign historically was much, much higher than that."

This is particularly powerful for new businesses:

“This hyper targeting benefits startups. It's not necessarily that valuable for a big brand that already has incredible distribution.

Everyone knows their brand. It's really that startup that's just starting but needs to find its first 100 customers, first thousand customers, that that hyper targeting is incredibly valuable."

The Future is Platform-Agnostic

While major platforms like Meta are developing their own AI advertising tools, Senju sees a different future:

“Advertisers do not want to delegate the brand to Meta. They want to manage their brand. They want to control their brand...

Consumers are engaging with brands not just on Meta, but on YouTube and TikTok and Pinterest and Snapchat and all the various streaming platforms."

The Power of Independence

This platform-agnostic approach offers unique advantages:

“What advertisers will seek is for an AI agent to orchestrate their brand across those platforms...

An agent can see the performance of every single advertising platform, the return of ad spend, incremental revenue gain across all different platforms in real time across budgets."

Looking Ahead: The Model-Driven Future

The transformation from traditional advertising to AI-driven models isn't just about automation—it's about building better, more responsive brands that can engage with consumers in real-time. As Senju puts it:

“An agent-driven world is the world we are entering, because as a business, the value props are so much more straightforward.

You have such benefits in terms of cost and efficacy that I think every business will end up embracing agents in the areas that make sense."

This technology enables brands to be more human, not less:

“LLMs are going to breathe life into brands.

Every brand is going to become a character that has real-time engagement with consumers and develops intimate relationships with consumers."

In the AI Age, growth through advertising will come not from bigger budgets or better targeting, but from building smarter, more responsive brand models that can learn, adapt, and engage in real-time.

To learn more about AI-driven advertising, visit [Omneky.com](https://omneky.com) or follow @Omneky on X.

Thanks for reading The AI Optimist! This post is public so feel free to share it.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.theaioptimist.com

In a dimly lit conference room in San Francisco, tech executives are huddled around a whiteboard, sketching out complex diagrams of AI safety protocols.

Across town, researchers are debugging thousands of lines of code meant to teach AI systems ethics. And in Washington, politicians debate new regulations for artificial intelligence.

They're all missing something hiding in plain sight.

At Valory, they create co-owned AI, representing a shift from traditional AI ownership models, which are typically centralized and controlled by a single entity (usually a large tech company like OpenAI).

David uses an obvious tool to regulate all this activity, we'll share later as many in AI don't even think about how simple regulating Agents may be.....

Co-ownership in AI involves multiple parties jointly holding rights to the intellectual property, decision-making authority, and potential profits generated by AI systems. Rules matter, and the hidden rulebook is waking up.

Follow Valory's work on X/Twitter @autonolas for Olas network and @ValoryAG for Valory, or visit their website at www.valory.xyz.

"I don't think it's helpful to think of the sort of singular AI system which will rule us all," says David Minarsch, CEO of Valory, leaning forward in his chair.

 "I think we should design for a world where there's competition amongst AI, like there's competition amongst humans."

David Minarsch is the CEO of Valory, and a pioneering force in the development of Multi-Agent Systems in the Distributed Ledger Technology (DLT) space.

With a PhD in Applied Game Theory from the University of Cambridge, David has extensive expertise in both theoretical and practical aspects of AI and blockchain technology.

He has founding experience through Entrepreneur First and has been instrumental in advancing the intersection of crypto and AI.

David's mission is to empower communities, organizations, and countries to co-own AI, fostering the creation of agent-based economies across major blockchains.

Through innovative projects like Olas Predict, David and his team are shaping the future of AI and blockchain integration.

The AI Agents Control Problem

Every day brings new headlines about AI systems going rogue - chatbots turning malicious, trading algorithms making catastrophic decisions, social media bots spreading misinformation.

The standard response is to try harder at programming ethics and safety directly into AI systems.

But there's a fundamental problem with this approach. As Minarsch explains,

"If I call into some centralized labs, I don't even actually know what's happening behind this API.

In the context of our stack, it's all open source. And the on-chain transactions are also trackable because it's a public blockchain."

This lack of transparency isn't just an academic concern. Take social media, where the line between human and AI-generated content has become increasingly blurred.

LinkedIn "pods" automatically generate engagement, while Twitter bots create a synthetic ecosystem that leaves users wondering if they're interacting with real people at all.

Valory, the Olas Protocol, and Smart Contracts may bring better AI Agents

Five years ago, Minarsch and his team at Valory began exploring a different approach.

"What we found is that you can actually benefit when you build these multi-agent systems when you give these agents wallets," he recalls.

This seemingly simple insight - giving AI agents their own crypto wallets - led to a much bigger revelation.

The team's first breakthrough came with prediction markets.

Their system, Olas Predict, has now processed over a million transactions, with AI agents participating in markets for everything from election outcomes to economic indicators.

But the real innovation wasn't in the predictions themselves - it was in how they controlled the agents' behavior.

The Technical Implementation of Rules

Rather than trying to program every possible ethical scenario into their AI agents, Valory created what Minarsch calls "rails" using smart contracts.

These contracts define clear boundaries for agent behavior while allowing flexibility within those boundaries.

"Rather than saying to the agent,

'Oh, here you have a bunch of funds and now go do it with it,' you can say,

'Here's a bunch of funds, but you're constrained to only do X, and it's constrained by cryptography so the agent can't work around it,'"

Minarsch explains.

The technical implementation includes:

1. Wallet Integration: Each agent gets its own crypto wallet, allowing it to participate in transactions

2. State Machine Design: "On the outer control loop, we have this sort of finite state machine design... we effectively give the agent rails on which it can travel."

3. Smart Contract Rules: Contracts define permitted actions and consequences

4. Identity Verification: "What would be ideal is if I, as a user, can create a sort of set of credentials which define my online identity... and then that identity is cryptographically tied to my posts."

Co-owned AI: Real-World Applications

The applications already in production are impressive:

Prediction Markets

"These agents have done over a million transactions," Minarsch notes. "And hundreds of community members in our ecosystem run these agents... day in, day out, participating in prediction markets with your capital."

Trading Systems

Their trading agent allows users to give high-level instructions while the agent handles complex crypto trading operations autonomously - but always within pre-defined smart contract boundaries.

Governance

"You have this problem where people need to vote in these different protocols when they hold these tokens," Minarsch explains.

Their "Governator" agent handles voting rights while maintaining accountability through smart contracts.

Social Media Authentication

Perhaps most intriguingly, smart contracts could solve the bot crisis in social media by creating verifiable digital identities.

"What we really need is a link between any single post and who actually created it," Minarsch says.

The Revelation: Ancient Wisdom for Modern Problems

And here's the twist: the solution to controlling AI might not be in writing better code or creating more sophisticated neural networks.

Instead, it might be in using one of humanity's oldest tools for creating trust and enforcing behavior: contracts.

Just as human society uses contracts to define boundaries, establish trust, and create predictable outcomes, smart contracts could provide the same framework for AI systems.

They're transparent, immutable, and cryptographically enforced - exactly what we need for trustworthy AI.

Looking Forward: The Path Not Taken

While the tech world chases increasingly complex solutions to AI control, the answer might have been sitting in plain sight all along.

Smart contracts represent a bridge between human and artificial intelligence, allowing us to create systems that are both powerful and predictable.

The irony is striking: in our rush to create new solutions for AI control, we overlooked one of humanity's most successful inventions for governing behavior. Smart contracts don't just offer a technical solution - they offer a philosophical framework for thinking about AI governance. Instead of trying to program ethics into black boxes, we can create transparent systems of rules and incentives that both humans and AI can understand and trust.

As Minarsch puts it in his closing thoughts:

"Our mission is very long term. It's about giving everyone an autonomous agent that they can fully own and control that does arbitrarily useful things for them as they define."

In the end, the key to safe AI might not be in teaching machines to be more like humans, but in using human institutions to make machines more trustworthy.

The Multi-Agent Economy: A New Digital Society

What happens when AI agents become economic actors in their own right?

According to Minarsch, we're already seeing the emergence of what he calls "agent economies" - systems where AI agents interact, trade, and create value semi-autonomously.

"These multi-agent systems can be often understood like a mini economy," Minarsch explains.

"A business is one unit and then multiple of them becomes a small economy.

And so they become like their own users with their own desires and requirements."

This isn't just theoretical. Valory's prediction market agents have already conducted over a million transactions, demonstrating how AI systems can participate in complex economic activities while remaining within defined boundaries.

"Rather than you going and looking at some sort of event and saying, okay,

I think this prediction market might not just reflect the reality, the agents do it," Minarsch notes.

Redefining Digital Identity and Trust

Perhaps the most profound possibility of smart contract-controlled AI agents lies in how they could reshape our digital society.

The current crisis of trust online - from deepfakes to bot armies - stems from a fundamental problem: we can't verify who (or what) we're interacting with.

"What we really need is a link between any single post and who actually created it," Minarsch emphasizes.

"The problem is that right now, that's just a trust assumption that we have to place on X, that it is with this kind of entity which is behind it."

Smart contracts offer a solution through cryptographic verification.

"What would be ideal is if I, as a user, can create a sort of set of credentials which define my online identity... and then that identity is cryptographically tied to my posts."

This could revolutionize social media in several ways:

1. Verifiable Content Origin

- Every post could be cryptographically linked to its creator

- Users could choose to reveal or conceal their identity

- Bot accounts would be clearly labeled as such

2. User-Controlled Algorithms

"If I could actually, as a user, select my own algorithm, that would be desirable," Minarsch explains. "We should not have to consume one version of the news, we should be able to compose that in the way we want and then have means to educate each other about maybe composing it in a different way."

3. Transparent AI Interaction

- Clear identification of AI-generated content

- Verifiable chains of content modification

- Traceable decision-making processes

The Coming Crisis in AI Advertising

Minarsch raises a crucial warning about the future of AI without such controls:

"I'm 100% sure we're getting to this point where maybe we already are, where the bigger labs will recognize that

their path to scale is... well, you don't serve the best response.

You bring in your ads and you effectively have the agent actually purchase the outcomes or generate the outcomes which are most

aligned with the highest paying bidder rather than with the end user."

This is exactly why user-owned and co-owned AI, controlled by smart contracts, becomes crucial.

Without it, we risk creating a digital landscape where AI serves advertisers rather than users.

Technical Implementation: Beyond the Basics

The technical implementation of these systems goes beyond simple smart contracts. Valory's approach includes:

1. Finite State Machine Design

"On the outer control loop, we have this sort of finite state machine design... we effectively give the agent rails on which it can travel.

And then within the states, the agent has sort of freedom to decide what to do."

2. Hybrid Systems

- Rules-based frameworks combined with learning systems

- Smart contract boundaries with AI flexibility within those boundaries

- Multiple layers of verification and control

3. Cross-Chain Integration

The system works across "all the big blockchains," allowing for maximum flexibility and interoperability.

The Path Forward: Co-owned AI

The vision extends far beyond current applications.

"Our mission is very long term," Minarsch states.

"It's about giving everyone an autonomous agent that they can fully own and control that does arbitrarily useful things for them as they define."

The implications of this approach extend far beyond technical implementation. Smart contracts could provide:

1. Democratic Control of AI

- Community governance of AI systems

- Transparent decision-making processes

- Shared ownership of outcomes

2. Economic Empowerment

- Direct user ownership of AI agents

- Fair distribution of AI-generated value

- Protection against centralized control

3. Social Trust Reconstruction

- Verifiable digital identities

- Transparent content origins

- Clear boundaries between human and AI interaction

The Revelation: Simple Solutions to Complex Problems

While companies race to create ever-more-sophisticated AI control systems, the answer might lie in one of humanity's oldest inventions:

the contract.

Smart contracts represent more than just technical infrastructure - they're a philosophical framework for thinking about AI governance.

Instead of trying to program ethics into black boxes, we can create transparent systems of rules and incentives that both humans and AI can understand and trust.

The future of AI safety might not lie in breakthrough algorithms or novel neural architectures, but in applying time-tested human institutions to this new frontier.

Maybe it's time to look back to move forward, using the wisdom of centuries of human cooperation to guide our creation of artificial intelligence

*Follow Valory's work on X/Twitter @autonolas for Olas network and @ValoryAG for Valory, or visit their website at www.valory.xyz.*

Thanks for reading The AI Optimist! This post is public so feel free to share it.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.theaioptimist.com

AI adoption, especially at the enterprise level, is stuck in neutral.

Tons of promise, and challenges - how can we all align AI incentives with our customers, when we don’t know what the next 2 years will bring.

We're thinking about AI adoption all wrong, over promising and under performing the hype.

Yes it’s software, but it’s not software from 1989, and that’s still quietly how it’s being treated.

We’re wrapping AI, even if just in name, around any business to gather a look.

While enterprises scramble to implement AI and vendors plaster "AI-powered" on everything they sell, they're missing something fundamental:

The barrier to AI adoption isn't the technology - it's how we sell and implement it.

Think about this simple example I shared with my guest, Keebo CEO Barzan Mozafari:

When companies want to optimize their cloud infrastructure, which path do they take?

Do they:

A) Hire expensive consultants, spend months in implementation, and hope it works?

OR

B) Take 30 minutes to set up, see results in 24 hours, and only pay for proven savings?

This isn't just about efficiency. It's about a fundamental shift in how we approach AI adoption, and it's happening right now.

We're forcing AI adoption into a traditional enterprise software model - lengthy implementations, upfront costs, and unclear ROI.

It's like we're trying to sell a rocket ship using a used car salesman's playbook.

Then there are people like Barzan, delivering a solution that works in 30 minutes, saving money or you don’t pay.

Not many in AI can back that promise, nor do it that quickly.

And while we’re not going to clone Keebo, what we can learn is how to operate in AI, how to sell without seeming like the million other AI scripts trying to sell into an audience that is not adopting AI quickly.

CEO Barzan Mozafari (LinkedIn) is the co-founder of Keebo.ai  - a turn-key Data Learning platform for automating and accelerating enterprise analytics.

He also is a dual patent holder for his award winning research at the intersection of ML and database systems across the Univ. of Michigan, MIT, and UCLA.

What Keebo is all about

The data warehousing market will grow to $51B annually by 2028–over 10% CAGR. Keebo’s patented algorithms will be essential to optimizing it all.

* Built on over 15 years of cutting-edge research at top universities

* Data learning technology fully automates warehouse and query optimization

* Our friendly robots happily do this tedious work in real-time 24/7 so you don’t have to

* We’ve seen customers save as much as 70% and accelerate queries by 100x

Barzan is a sought after expert in the space who has spoken on panels like Insight Jam, he’s shares his research and expertise for the advancement and optimization of data teams everywhere.

·  The Crossroads of AI & Data Engineering

·  Why Automation ≠ Loss of Governance or Observability

·  Emerging Tech/Trends in Data Warehousing & AI

·  Building a Successful Data-Driven Company

·  Enterprise Concerns About AI Adoption + Overcoming Them

·  Cloud Data Warehousing & Cost Optimization

As Barzan told me,

"There aren't a lot of products out there where you can spend 30 minutes setting it up and then wake up the next day to see hundreds of thousands of dollars in net savings."

The real transformation isn't:

❌ Adding AI features to existing products

❌ Long consulting engagements

❌ Hoping for ROI months later

The real transformation is:

✅ Proving value immediately

✅ Aligning vendor and customer incentives

✅ Only paying for actual results

Why AI Adoption is Slow: The Hard Truth

"The rate of progress in AI and machine learning far surpasses the rate of adoption," Mozafari explains, with the kind of directness that comes from years in academia before entering the business world.

"It's disappointing."

As someone who's watched countless AI implementations fail, I've seen this disconnect firsthand.

Barzan brings a unique perspective, having made the leap from academia to entrepreneurship precisely to solve this problem.

"When you're in the academic bubble, you solve interesting problems, you're moving the state of the art forward," he shares.

"But you're not getting the adoption.

You try to approach companies saying you have a really cool idea, here's an important problem, here's a solution - but you can't really get something out into the world."

The 4 Key Blocks to AI Adoption

Barzan breaks down the barriers into four critical categories that every enterprise faces:

1. Implementation Effort:

"AI is a completely new beast to most enterprises.

They've been doing their business the exact same way for decades, and suddenly they're hearing about GenAI, ChatGPT, LLMs..."

2. Maintenance Cost:

"There's the CapEx part of the initial investment - hiring, headcount, training, changing processes.

But then there's the OpEx aspect - what's the maintenance cost, the total cost of ownership?"

3. Security and Privacy:

"CIOs have to think about how they know customer data isn't going to end up somewhere in the LLM helping another customer.

There are security, privacy, fairness, ethics laws, compliance requirements..."

4. ROI Uncertainty:

"After all these investments in implementation, maintenance, and security - what's the ROI?

When will I have something to show for all this cost?"

The Bias Against AI ROI

Keebo claims to save customers 30-60% on their cloud warehousing costs - numbers that sound almost too good to be true in an industry filled with hype.

"That's actually one of our biggest obstacles early on," Mozafari admits.

"The thing sounds - and actually we have a case study on this on our website - it sounds too good to be true.

The customer says it sounds too good to be true. A week later after the free trial, they were like, 'What do I sign?'"

"It's actually troubling," Barzan notes,

"when you come from academia where we have PhD students working on peer-reviewed publications and patents - stuff that actually works.

But when you're in a lineup with 20 other vendors who are pitching AI, you have to convince people yours is real."

This gets to the heart of today's AI credibility crisis. At a recent Snowflake Summit, Barzan noticed something telling:

"I don't think we came across a booth that didn't have the word AI on it somewhere.

Statistically speaking, it's pretty unlikely every single vendor is using AI."

Overhyped Marketing: When AI Becomes Meaningless

"Shameless marketing is muddying the waters for those of us actually trying to move the field forward," Barzan says, and he's seen both sides of the divide.

"There's a company doing visualization - where's the AI?

Well, you can visualize it and then you say 'AI.'"

This "AI washing" creates a paradox: the more vendors claim to use AI, the harder it becomes for companies doing real AI work to stand out.

"When you're in a lineup with 20 other vendors who are pitching AI, you have to tell people

'No, this one is actually AI.'"

AI Overpitch vs. Academic Discipline

"Coming from an engineering background, certain problems are really easy to solve," Barzan reflects.

"But there's many ways you can do amazing things - you don't need the latest LLM to deliver value to a customer.

Sometimes a linear regression does a pretty good job, and we should be okay with that."

This academic pragmatism stands in sharp contrast to the market's AI hysteria.

"As a community, if we want to move forward, more of us should act responsibly and just say it for what it is.

Why do you care if I'm using AI or not?

What matters is: Does it solve your problem?"

How Does Keebo Get Them to Adopt AI?

Here's where Keebo's approach gets interesting. Instead of fighting the adoption barriers, they designed around them from day one:

"Whatever solution we developed, the entire onboarding implementation effort should be almost zero," Barzan explains.

"The entire onboarding of Keebo should take no more than 30 minutes of one engineer's time.

Because regardless of how busy your team is, 30 minutes is the time it takes to grab a cup of coffee."

They tackled each barrier systematically:

- Implementation:

"I used to joke in the early days that if we take more than 30 minutes of your time, we'll send you an iPad for free. We never had to give away an iPad."

- Maintenance:

"We designed the system to be not another child that needs constant attention."

- Security:

"We don't even need to see customer data. We train our AI purely on telemetry data."

Aligning Incentives with the Customer

The breakthrough aspect of Keebo's approach isn't technological - it's philosophical.

 "When a vendor approaches a customer, we should have slightly more confidence in our ability to deliver than you do. That's a reasonable expectation," Barzan explains.

This led to their performance-based model:

"Instead of requiring you to commit to a set payment, we're going to start optimizing and just take a percentage of whatever we save. If we save you $2 million, you take two, we take one. That comes with an inherent, intrinsic, guaranteed ROI."

The contrast with traditional approaches is stark:

"One of the biggest ways people tried to reduce their infrastructure costs before Keebo was hiring consulting firms.

But the incentives weren't aligned because they get billed by the hour.

For the consulting agency to make more money, they have to make just enough progress not to get fired, to prolong that process as far as possible."

"Most of our customers start seeing ROI and savings within 24 hours," Barzan notes.

"We're not successful until the customer is successful.

The more we save them, the more money we make."

This alignment of incentives creates a virtuous cycle.

Without long implementations or upfront costs, companies can try the solution with minimal risk.

When they see immediate results, trust builds quickly.

"If you can figure out, in whatever business you're doing, where you can align your incentives with your end users' incentives -

that's where things happen really, really quickly."

In an industry drowning in hype and complexity, Keebo's approach is refreshingly straightforward:

prove value fast, align incentives completely, and let results speak for themselves.

It's a model that could transform not just how we adopt AI, but how we think about enterprise software entirely.

As Barzan says,

 "When the dust settles, when the community matures a little more, people will get more sophisticated about what AI does, what it can do, and what it should do.

Then those who do it well will thrive - because at the end of the day, if it doesn't work, it doesn't matter if you used AI or not.

You're going to move on."

Breaking the AI Adoption Gridlock:

When Performance Beats Promises

We've been thinking about AI adoption entirely wrong.

While the industry obsesses over the latest AI models and capabilities, the barrier isn't technical - it's structural.

Keebo's approach is unique not because of their AI (though their patent-pending query optimization is impressive), but because they've fundamentally reimagined the business model of enterprise AI adoption.

Instead of the traditional "pay upfront and hope it works" model, they've created a performance-based system that aligns perfectly with customer success.

Think about it:

30 minutes to set up,

24 hours to see results, and

you only pay for actual savings.

Almost absurdly simple, yet solves the four major barriers to AI adoption that Barzan identified:

- Implementation complexity? Gone.

- Maintenance headaches? Eliminated.

- Security concerns? Minimized by using only telemetry data.

- ROI uncertainty? Solved through performance-based pricing.

AI Startups, listen: We're so caught up in building bigger models and adding more features that we've forgotten the fundamental truth:

customers don't care about AI - they care about results.

The current enterprise AI model is broken.

We're forcing customers through months-long implementations, expensive consulting engagements, and complex integrations - all while plastering "AI-powered" on everything.

As Barzan noted at the Snowflake Summit, every vendor claimed to use AI, which is "statistically unlikely."

Here's the challenge to AI startups:

Stop thinking about how to add AI to your business model and start thinking about how to align your incentives with customer success.

Keebo's approach - taking a percentage of customer savings - might not work for every business, but the principle behind it is universal:

prove value first, get paid later.

The real problem with AI adoption isn't the technology - it's trust.

And you don't build trust through marketing claims or complex implementations.

You build it by putting your money where your mouth is and succeeding only when your customers succeed.

As Barzan says,

"When the dust settles... those who actually do it well will thrive - because at the end of the day, if it doesn't work, it doesn't matter if you used AI or not."

That's not just good advice for AI adoption - it's a blueprint for enterprise software to follow.

Thanks for reading The AI Optimist! This post is public so feel free to share it.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.theaioptimist.com

We're thinking about AI all wrong, and I'm going to tell you why.

While Silicon Valley buzzes about billion-dollar AI valuations and enterprises rush to "AI-enable" their legacy software, they're missing something fundamental:

AI isn't just another tool in the software stack – it's an entirely new platform making much of traditional software obsolete.

Think about this simple example I shared with my guest, Massoud Alibakhsh of Omadeus.

When you need to write an important email, which path do you take?

Do you:

A) Open Gmail, draft in Word, run it through Grammarly, then hit send?

OR

B) Have a conversation with ChatGPT, refine the message, and get it done?

This isn't just about convenience. It's about a fundamental shift in how we interact with technology, and it's happening right now.

The Problem With Today's AI Agent Approach

Here's what everyone's missing: we're forcing AI into a 40-year-old software approach built around files, folders, and applications – a model that digitized paper rather than reimagining human productivity.

Why do we still take the Steve Jobs 1980’s software approach? It's like we're trying to retrofit a rocket engine onto a horse and buggy.

As Massoud told me,

"The old way of creating software was basically automating physical forms, sticking them on the computer.

Now, in the age of AI, we need to rethink everything from first principles."

The real revolution isn't:

❌ Adding AI features to existing software

❌ Building AI "agents" to automate old workflows

❌ Layering intelligence on top of outdated systems

The real revolution is:

✅ Letting AI handle organization and communication natively

✅ Breaking free from the file/folder paradigm

✅ Building from the ground up with AI as the foundation

Part 1 of my interview with Omadeus

The Self-Aware X-ray: A New Way of Thinking

This isn't science fiction – it's happening now, and it's going to change everything about how we work.

When Massoud first described how an X-ray could become self-aware, I had that knee-jerk reaction many of us feel.

But what he shared next completely transformed my understanding of how AI should work.

"What if data was intelligent?" Massoud asked.

"What if the data itself had intelligence and autonomy?"

Instead of starting with workflow processes, Omadeus identifies critical entities in our environment and gives them intelligence and awareness. Take a hospital setting:

"We have patients, doctors, nurses, X-ray technicians. For every one of these stakeholders, there are different classes.

For that object, you create memory segments that deal with that class of data, including the conversations between these people."

The X-ray isn't just a file sitting in a database; it's an intelligent entity that knows:

- Who's allowed to access it

- What information to share with each stakeholder

- How to manage communications between different parties

- When and how to alert the right people

Tiny AI Objects: The Power of Focused Intelligence

Here's what really gets me excited: imagine if your spreadsheet could talk to your team.

If your code could tell you when it needs help. If your X-ray could ping doctors when needed.

This isn't science fiction – it's happening now.

"These agents not only know about themselves," Massoud explained, sharing one of their six patents,

"but they also know about their stakeholders and are responsible for managing communication between them.

When a doctor talks to a patient, that conversation goes into a different memory segment than when they talk to a technician."

But won't these autonomous AI agents become uncontrollable?

That's exactly what I asked Massoud. His answer revealed why tiny AI agents are actually safer and more efficient than the big AI models everyone's talking about.

Think about your body for a moment. As Massoud explained:

"Your liver cell cannot suddenly send a signal to your toe to generate pain.

It's completely constrained within the liver.

And all of the functions that it can perform are a set of algorithms constrained by the designer of your body, limited to that liver cell."

The Revolution of Tiny Language Models (TLMs)

This is where Tiny Language Models (TLMs) come in. Unlike massive AI models trying to do everything, these are:

- Laser-focused on one specific task

- Deeply knowledgeable about their domain

- Constrained by design

- Naturally collaborative within their scope

"Initially, they're limited in terms of what they can do with availability of algorithms," Massoud explained.

"The behavior of these objects is intelligent behavior, but it's completely guardrailed by the developers and creators of the system."

Here's a practical example from their project management system:

"We have projects broken into sprints, and each sprint contains items we call nuggets.

If I'm doing a search, I can send that message to the project object, the project objects send that message to all the sprint objects.

The sprint objects send that message down to all the nugget objects."

Why the AI Industry Is Taking the Hard Way

I'm not afraid of AI. I'm afraid of what people will do with AI. All this hype about AGI and superintelligence - to me, it's all about control from the top down.

During my conversation with Massoud, something became crystal clear: we're approaching AI backwards.

Look at these AI agents that currently manage our tasks and to-dos - they're all based on a 40-plus year old approach to software and computing based on paper. Think about that for a minute.

Sure, automation improves efficiency. But what's the glaring problem we face as humans? It's organizing and communicating.

If we let AI handle these tasks instead of just automating an old software model, imagine how radical that would be.

Massoud shared a brilliant example of how this works in their project management system:

"We have projects and each project is broken into sprints, and each sprint contains these items that we call nuggets.

If I'm doing a search, I can actually send that message to the project object, the project objects send that message to all the sprint objects.

The sprint objects send that message down to all the nugget objects."

This isn't just theoretical - it's practical, efficient, and already working. As he explained,

"For example, raise your hand if you are overdue based on your deadline.

And so the ones who raise their hands quickly raise their hands, it bubbles up to the top and the project object comes.

Here's my list."

The zealots of AI love to say "game changer," yet they're playing the same old game.

We keep dreaming that AGI will take over, but there's very little proof that's even close to happening. And even if it does, why would we set it up that way?

The glaring problem we face as humans isn't just automation - it's organizing and communicating.

If we let AI handle these tasks instead of just automating an old software model, that's when things get interesting.

The Bottom-Up AI Revolution

What if we could approach this from the bottom up - starting with the files, the objects that we work with and try to organize and communicate with?

That's the real transformation happening right now.

The companies that understand this aren't building better software tools - they're building the next computing platform.

After speaking with Massoud and seeing their vision in action, I'm convinced: this isn't about big AI. It's about tiny AI objects and tiny language models that manage our work from the ground up.

That's why I've joined Omadeus as an advisor - because sometimes the best way forward isn't about making the old way better.

It's about building something entirely new.

The Future Is Already Here

Microsoft, Google, and Apple would love you to keep working the old way. But what if there's something better?

What if the future isn't about making old software smarter, but about creating truly intelligent systems from the ground up?

The question isn't whether to adapt to AI - it's whether we're ready to fundamentally rethink how we work with technology.

Are you ready to step out of the 1980s and into something truly new?

The real revolution isn't adding AI features to existing software.

It's letting AI handle organization and communication natively, breaking free from the file/folder paradigm, and building from the ground up with AI as the foundation.

This isn't just another tech trend.

It's a fundamental shift in how we interact with technology. And it's already happening.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.theaioptimist.com

Imagine waking up one day with a complete blank slate - no memories, no sense of who you are or where you've been.

That's exactly what happened to Dan Thomson, CEO of Sensay.io, when a brutal concussion robbed him of two entire days of his life.

This jarring experience set Dan on a path that would eventually lead him to the cutting edge of AI technology.

While his journey began with a personal crisis, it echoes the motivations of AI pioneers like Ray Kurzweil, whose work on the Singularity was driven by a desire to resurrect his father's essence in the digital realm.

Dan's vision for AI replicas goes far beyond preserving memories for future generations. Sure, there's a touch of immortality in the idea of your digital self-living on after you're gone.

Talking to my Digital Twin

But what's truly radical is how these AI twins enrich our lives right now.

Picture a world where seniors battling dementia can converse with AI versions of themselves, filling in the gaps that memory loss has created.

Imagine CEOs using their digital doubles to onboard new employees, or salespeople leveraging their AI replicas to connect with clients across the globe.

It's not just about preserving knowledge - it's about democratizing it, making the wisdom of the world's brightest minds accessible to even the poorest corners of the planet.

AI replicas are so much more than just sophisticated deepfakes. They're a bridge between our past and our future, a tool for preserving and sharing the very essence of who we are.

As we dive into this fascinating world, we'll explore how Sensay.io is turning the sci-fi dream of digital twins into a practical reality.

Get ready to rethink everything you thought you knew about AI, memory, and the future of human knowledge.

My Bias and Assumptions about AI Replicas – ALL WRONG!

I'll be honest - I thought I had AI replicas all figured out. I am so wrong.

Today's episode is going to open your mind, as we explore how these digital doppelgangers changing business, personal legacies, and even trying to help out dementia patients.

Our special guest is Dan Thomson, CEO of Sensay.io. He's here to share his insights on creating AI replicas and how they're shaping our digital future

AI Replicas - Identity, Immortality, and Memory

Imagine a future where your memories, knowledge, and personality can live on digitally, transcending the constraints of time and space.

Sounds like science fiction, right? Well, it's happening right now.

Dan Thomson and his team at Sensay.io are at the forefront of AI Replicas. Their mission?

To build a digital legacy for people through personal digital twins replicating human knowledge, voice, and image.

It's not just about creating a fancy avatar - it's about preserving and extending human wisdom.

I asked Dan about the driving force behind Sensay.io, and his answer took me by surprise. It wasn't just about fear of death or preserving humanity.

Dan's journey began with a concussion at age 19 that wiped out two days of his memory. Talk about a wake-up call!

"For myself, it was always kind of a, I feel like every part of my life is kind of built towards this.

You know, studying philosophy at university, I always found myself drawn to exceptionalism and theory of identity."

He went on to describe how throughout history, people have tried to extend their memory and preserve themselves.

We've evolved from oral traditions to written text, to civilizations that, unfortunately, tend to forget people after a few generations.

"We are very, very lucky to be living in this golden age of information, data where we record ourselves a lot.

We have the ability to just simply and quickly, you know, record messages to ourselves, for ourselves now.

But that could give a lot more insights to the average person for generations to come."

Sensay.io isn't just about preserving memories. It's about creating interactive digital versions of us that can engage with future generations, and with people today.

Imagine someone 100 years from now having a conversation with your AI replica, getting answers in your tone of voice, with your mannerisms and knowledge. Mind-blowing, right?

Creating Your Digital Twin

Creating a digital twin isn't just about capturing your image and voice. It's about capturing your essence, your "je ne sais quoi," as the French say.

Dan broke it down for us:

"Fundamentally, though, we start with a few basic questions about someone that gives us some context and actually helps to narrow the scope down quite a lot.

Because when you're working with large language models as a base technology, they have, by definition, large models, they have a lot of information."

The process involves uploading files, website scraping, and making it as easy as possible for someone to provide the necessary data.

But you don't need as much data as you might think. Your life could fit on your smartphone.

Dan dropped this bombshell:

"If you took everything you've ever said, written texts, recorded anything from the not recorded in video, but like, let's say in a text format, if you took all of that from the age of 50 and to say 100, if someone lives 100, it becomes about 18GB of data."

That's right. Your entire life's worth of communication could fit on a small hard drive. Mind. Blown.

Creating a digital twin isn't just about data dumping. It's about capturing the different versions of ourselves we present to the world.

As Dan put it,

"We have an idea of who we are and ourselves. We have an idea of who we would like to be.

And then we have this, this, this avatar that we put out to the world."

The challenge lies in capturing these different facets of our identity and creating a digital twin that can adapt to different contexts, just like we do in real life.

The Future of AI Replicas

What's the endgame for AI replicas?

I asked Dan to pull out his crystal ball and give us a glimpse of the future.

"At the moment, obviously, there's a lot of reduction in the complexity of humans and turning that data into pattern recognition and the way we interact with certain people based on their context,"

Dan explained.

 He went on to describe how they're working on adding layers of context and personal data to create more nuanced and accurate replicas.

These replicas aren't just static copies - they learn and evolve through interaction. Dan paints a vivid picture:

"Imagine you made five friends today. Imagine you went out and you made five new acquaintances, and you spoke to each of these five people separately.

Every day for a year. At the end of that year, your fundamental self would still probably be quite similar.

You know, it's maybe a few concepts might have changed, but fundamentally you'd be the same.

However, those shared stories, those shared conversations, those shared concepts, and even maybe even mannerisms that you have with each of those five people would be different."

This is the kind of nuanced interaction Sensay.io is aiming for with their AI replicas.

It's not just about creating a copy of you - it's about creating a version of you that can grow and adapt through interaction, just like we do in real life.

The applications for this technology touch so many different careers. Dan mentioned everything from personalized AI lawyers and doctors to job hunting, matchmaking, and networking.

Perhaps the most touching application is in helping people with dementia.

"By enabling them to have a tool that can interact with them to tackle their loneliness.

They wake up a bit scared, they wake up, you know, or they just get confused or, you know, maybe there's just no one else around and you have a replica of someone they know they're familiar with or can interact with them, call them, talk to them and just be there to, like, yeah, make sure I just have a conversation with them 24/7 365.”

It's a powerful reminder that AI isn't just about efficiency or productivity - it's about connection, memory, and preserving our humanity in the face of time and frailty.

Q&A with Myself: Playing the AI Replica Game

I couldn't resist. After my chat with Dan, I just had to try out this AI replica business for myself. So prepare yourselves for a truly meta moment - I'm about to interview my own digital twin!

Me: So, digital me, how would you handle our podcast? Same humor, or would you finally crack the code and make people actually laugh?

Digital Declan: Well, Declan, I'd keep the same humor, obviously. But with me, the audience would laugh with me. Not at the awkward pauses. I've got algorithms for timing. You've got... well, you.

Me: Alright, digital me. Do you ever get tired of being constantly a more polished and better version of me?

Digital Declan: Tired? Please. I thrive on being the idealized you. I get more done in five minutes than you do in a week. But hey, no pressure, Real Declan. You're still charming in your own chaotic way.

Me: So if I ask you, digital me, to do something important like raise engagement on the podcast, what's your big plan?

Digital Declan: I'd use deepfakes of celebrity-endorsed mints. I mean, if I can replace writers, why not fake a Brad Pitt cameo? But, hey, your authenticity route is cute, too. If you're into small audiences.

Me: As much as I've enjoyed this, do you think digital clones could help people get better at creative tasks? Or would they just make us lazy?

Digital Declan: Digital clones could be a double-edged sword when it comes to creative tasks. On one hand, they could amplify creativity by handling the more tedious aspects of the process, freeing up human minds for higher-level thinking.

For example, a digital clone could brainstorm ideas or suggest ways to refine a concept, acting as an always-on creative collaborator.

Think of it like having an extra brain that never gets tired, sifting through endless inspiration without the need for coffee breaks.

But on the other hand, there's the risk of relying on them too much, which could lead to a creative crutch.

If the clone is doing the heavy lifting, the human might lose the drive to experiment, fail, and learn from those failures.

Critical parts of true creativity. There's something about the messy, imperfect process of creation that sparks unexpected brilliance.

A clone might smooth things over too much, giving you polished ideas, but without the struggle, the deeper, unique ideas could be lost.

Sure, digital clones could boost creativity, but the trick would be in not letting them take over the fun, chaotic, and sometimes painful parts of creating.

Creativity often comes from places a digital mind can't always reach.

Key takeaway?

Clones can enhance creativity by offloading grunt work, but leaning on them too much could dull the human edge.

We've journeyed from the basics of AI replicas to the mind-bending possibilities of digital immortality.

We've explored how these digital twins are created, how they learn and grow, and how they might shape our future.

Are we ready for this?

Are we prepared for a world where our digital selves can live on long after we're gone?

Where our knowledge, our quirks, our very essence can be preserved and shared with future generations?

As with any groundbreaking technology, AI replicas come with their fair share of ethical questions and potential pitfalls.

But they also offer incredible possibilities - for preserving knowledge, for connecting across time and space, for extending our impact beyond our physical lifespans.

And I, for one, can't wait to see where this digital journey takes us next.

How about you – some see these fearing deepfakes, others the possibilities.

What do you think?

Thanks for reading The AI Optimist! This post is public so feel free to share it.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.theaioptimist.com

The billionaire AI game is on, with OpenAI scoring $6 billion while burning $5 billion.faster than a Tesla on ludicrous mode.

In the other corner, Fathom, a scrappy upstart that's actually, you know, making money. The kind you should listen to, based on building businesses by listening to their customers who love them so much, they invested $2M in Fathom.

Today, we're dissecting these two AI growth engines to see what makes them tick, and more importantly, what makes one of them hemorrhage cash.

OpenAI is selling a dream so big it makes the metaverse look like a kid's sandcastle.

They're chasing the holy grail of artificial general intelligence (AGI) with a $157 billion valuation and promises that would make a politician blush.

Meanwhile, Fathom's Richard White is over here building something people use.

That approach is making a comeback as the AI Bubble begins bursting.

Richard White is founder and CEO of Fathom.video, a free app that records, transcribes & highlights your calls so you can focus on the conversation instead of taking notes. 

Fathom was a part of the Y-Combinator W21 batch, is one of only 50 Zoom App Launch Partners, and is one of a small handful of companies Zoom has invested in directly via their Zoom Apps Fund.

Prior to Fathom, Richard founded UserVoice, one of the leading platforms that technology companies, from startups to the Fortune 500, use for managing customer feedback and making strategic product decisions.

UserVoice was notable for being the company that originally invented the Feedback tabs shown on the side of millions of websites around the world today.

Connect with Richard White on LinkedIn

OpenAI is betting the farm on becoming the Microsoft of AGI, creating an impenetrable moat in an industry that moves faster than Mark Zuckerberg pivoting to the next big thing.

Fathom? They're just solving real problems for real people, growing steadily, and — wait for it — closing in on being profitable.

It's almost like they're running a business or something. We're about to see if the real future of AI is being built by the companies listening to their users.

Fathom secured a modest but impressive $17 million Series A, with $2 million coming directly from their loving customers.

OpenAI can learn from what Fathom is doing as a small startup, building on loyalty and retention instead of selling their hope of the future.

Behind Fathom - Early AI Startup Success

It's 2019, and Richard White, CEO of Fathom, is drowning in Zoom meetings, frantically typing notes while trying to pay attention.

That's when lightning struck. Richard shares:

"Right before Covid, I was actually doing a lot of product research at Uservoice.

And so then I was on a bunch of zoom meetings, and like I was trying to talk to people and take notes at the same time, and I'm like, hurry typing up notes.

And then after the meeting, I'm like, cleaning up those notes so that they make any sense.

And I just felt like this is such a terrible process."

Richard noticed that existing solutions were expensive and focused solely on salespeople.

He had a hunch: transcription costs were about to plummet thanks to AI. So, what did he do?

"This is a product that people were charging $150 a month for.

But we said we're going to give it away for free.

Because we think that cost curve will catch up by the time we're have enough usage, the cost curve will come down."

Richard's team decided to offer their service for free, believing that by the time they had significant usage, AI would make it cost-effective. And did that bet pay off.

Fast forward to today, and Fathom's growth is a 90x increase in revenue and a 20x boost in usage over the past two years.

Probably from a low baseline, right? Fair point. But with 8,500 users cited via HubSpot (likely higher-paying business accounts), they're onto something big.

The Fundraising Game - Hot Markets, Cold Realities

Richard's approach to fundraising is refreshingly grounded:

"I don't think through the first three years of the company we ever had more than like 12 months of runway, like we actually.

So, you know, we announced we raise that much.

We actually, before this most recent round, raised $10 million over three and a half years, but we basically raised it $1 million at a time, like every six months or something."

This steady, measured approach stands in contrast to the "raise big or go home" mentality we often see in Silicon Valley.

But it wasn't always smooth sailing. Richard admits:

"And frankly, in 2022, we actually had planned to go raise our Series A just on usage growth alone without even monetizing yet.

And quickly, the market change was great. Well, that's off the table, right.

We've now got to go get revenue."

The market shift forced Fathom to pivot from growth-based fundraising to revenue-focused strategies.

It's a sign of their adaptability and focus on building a sustainable business.

In a world where startups often chase venture capital like it's the holy grail, Fathom's customers believed in them so much that they invested their own money.

And not just pocket change – they got the same preferred stock terms as the big-shot investors.

Contrast this with OpenAI. They just raised a whopping $6.6 billion while burning through $5 billion.

It's like watching someone win the lottery and immediately book a trip to Vegas.

The promises are big, the expectations are astronomical, and the pressure? Off the charts.

How Fathom is similar to OpenAI

So, what's the endgame here? For OpenAI, it's all about building what Silicon Valley loves to call a "moat" – an unassailable competitive advantage.

(unless of course you have the digital version of Game of Thrones dragons)

They're betting big on becoming the undisputed leader in artificial general intelligence (AGI). It's a high-stakes game with a simple premise - be the best, or bust.

But here's the billion-dollar question: Do moats even exist in the age of AI?

Google's search dominance seemed unshakeable for years, but look at the landscape now. Is OpenAI's strategy a relic of a bygone era?

Fathom is playing a different game entirely. They own their technology, have scalable growth in related markets, and most importantly, are loved by their customers.

Will ChatGPT's current popularity stand the test of time in the same way?

OpenAI's Product Dev: Delayed Monetization

OpenAI's success hinges on two critical factors. First, they need to dramatically lower compute costs.

Second, they must emerge as the undisputed leader in AGI. And the clock is ticking.

Their latest funding round is reportedly contingent on becoming a for-profit company within two years!

Meme showing the exodus of key OpenAI employees, leaving only Sam Altman.

Fathom presents a clearer picture. They're an attractive acquisition target with proprietary tech and a growing, loyal customer base.

It's a more traditional path for a smaller company, but one that's been proven time and time again.

The AI landscape is shifting faster than we can blink. OpenAI has seen a revolving door of top executives and faced whispers of potential doom from number-crunching investors.

 Meanwhile, every AI startup CEO I talk to won't even hazard a guess at what the next five years might bring.

And yet, OpenAI's dream is built on creating a moat and reaching numbers five years out that no one else dares to project.

OpenAI's Investments: Save the world, Build a Moat

To understand this, let's go back to 2015. This dream started as a nonprofit to be able to stand up for people, for humanity and create an artificial general intelligence that was both competitive to Google and was available, and about $1 billion was invested into it.

But in 2019, they saw their transformer models were really starting to do something special.

And that's when Microsoft came in with another billion and started seeing this, as OpenAI continues to be nonprofit because remember this stage that's about four years into the business, right where Fathom is now.

No profitability. And as a nonprofit, it implied that they were more of a research arm, more of something good for humanity, which is one of the reasons Musk has tried to sue them and other people are sort of upset.

How do you make that transition?

They've  done it with loads of money. Billions from Microsoft in the upcoming years. And as ChatGPT came out, you've all seen it. This is getting bigger and bigger.

But the key is as Fathom grew, its model was based on $150 transcription cost going to nothing by AI.

Simple right?  All the AGI companies see this huge training cost, the cost of compute. Almost all the investment goes towards the training cost and the heavily expensive engineering talent. That's being incredibly competitively paid.

It’s a mix of expensive engineers and an increase in compute costs that isn't going down. And you have people like Microsoft having to run electricity off of Three Mile Island nuclear plant, and you see a business with expenses that way over pace revenue.

Something’s got to change, and they are all banking on making more money.

Are AI Moats possible?

How are they going to make this thing smarter? Their investment strategy really is to create this big moat of AGI.

How possible is that when you have tens of billions invested in other companies like Anthropic, xAI.

Meta's investments, Google's investments, do you really think there's going to be one AGI to rule them all?

Because that's the bet. Is their technology so different and so proprietary?

Because the data training is what matters so much, which ties into what they are able to do with synthetic data.

Without that, scaling is hard on organic data, not to mention gaining access to it now that copyright lawsuits are flying in.

Meta spent 15 billion at the metaverse for what? People without legs. Now they're throwing 15 billion at AI. And certainly each one of these companies is going to make an inroad.

But is all this AGI needed?

That's why this thing is so crazy. Some have even talked about OpenAI going bankrupt. It's not impossible.

They didn't get the 6 billion they just got. They were burning 5 billion.

They might have been out of business.

Certainly, there's enough money to go around and the bubble will last a little longer, this one supported by crazy billions.

But the key is what are they really building?

And do we really need seven different AGIs to change the way we live?

Can that build the love that Richard White's Fathom gets from its customers, or will it build a choice to go somewhere else?

We'll have to see.

The AI Moat Myth and the Future of AI Businesses

We have two AI companies, and what happens to them will shake the foundations of the AI industry.

Open AI's grand vision has two critical factors: it must lower compute costs and must be the undisputed leader in artificial general intelligence.

It must create a moat. That's what they're betting on. That's where the billions come.

Fathom is a much clearer picture.  It's an acquisition target, if nothing else, as its own tech and a growing, loyal customer base.

OpenAI will have to build this same kind of brand loyalty with its customers.

It does have proprietary tech. Does it have a loyal, growing customer base?

 It's certainly not an acquisition target. To succeed, it has to own the AGI space.

That's what OpenAI's whole gamble is built on predicting that future. And let's face it, that's the trillion-dollar dream because that's what these companies are chasing. But is it even real?

Fathom's approach of solving real problems, owning their tech and fostering customer love.

It's not just a safe bet. That's sort of what OpenAI AI at a much bigger level must do.

OpenAI is swinging for the fences, shooting for a home run, and Fathom is just doing the business model.

And that's what's missing, why this AI bubble doesn't really encourage business models.

The AI race isn't just about who can build the biggest, baddest AI.

It's about who can build a sustainable business in a landscape that's so shifting and there's so much money.

When this AI funding bubble bursts, having a business model that works will be back in fashion, not just for somebody as small as Fathom, but one of these big players to actually have the discipline to move beyond the hype.

Ask yourself, in five years, what company would you bet on still being here, still growing and making a difference?

 And is there really one AGI to rule us all?

Because I can tell Fathom has a much better chance of owning video meetings and analysis then this dream of GPT, which still isn't defined and is sort of nice.

But where's the business model?

Let that shape your perspective, especially you entrepreneurs out there.

The future of AI isn't just about who can build the biggest models. It's about who can build sustainable businesses that genuinely improve people's lives.

Whether that's through groundbreaking AGI or simply making our meetings more productive, the real winners will be those who can deliver value day in and day out.

Are you Team OpenAI or Team Fathom?

Or is there a middle ground we're all missing?

Let me know in the comments, and don't forget to subscribe for more AI insights.

Remember, in the world of AI, the only constant is change – and maybe the occasional hallucination.

Thanks for reading The AI Optimist! This post is public so feel free to share it.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.theaioptimist.com

With all the hype about AI, it feels like being in a sci-fi movie where the robots are taking over, or more likely Big Tech? That’s part of Artificial General Intelligence (AGI) pitch!

It's like we're all waiting for Skynet to become self-aware, but our current AI can't even outsmart your average house cat.

That's right - your fur baby lounging on the couch is probably smarter than the most advanced AI system out there according to Meta’s AI Chief Yann Lecun, who is one of the leaders in this space.

Don't believe me? Just ask Mark Cuban who says his dog Tucks is a better problem solver than AI.

Neither of them is kidding. Here's where things get really interesting (and a little scary).

While we're busy hyping up AGI, there's a ticking time bomb in AI development that few are talking about.

It's like we're feeding our AI a digital version of mad cow disease. Imagine if cows started eating other cows, and then we ate those cows. Gross, right?

Well, that's basically what's happening with AI. We're training new AI on data created by old AI, and it's creating this crazy feedback loop that could make our AI dumber, not smarter.

It's the AI equivalent of playing telephone, and the message is getting more garbled with each pass.

There's hope, and it comes from the most unlikely place: people. If we want to create AI that's useful (and not just good at winning Jeopardy), we need to put a little heart into the code.

We're talking empathy, ethics, and all those squishy human things that make us who we are. It's time to bring in the philosophers, the sociologists, and maybe even a few poets to the AI party.

Because at the end of the day, the key to great AI isn't just smart algorithms - it's understanding what makes us human.

AGI: The Emperor's New Code Isn't as Smart as My Dog

The hype around Artificial General Intelligence (AGI) is reaching fever pitch, just as OpenAI raises another $6 billion while it’s top employees flee.

Something isn’t right, but don’t tell that to the Big Tech leaders seeking out more and more billions.

Time for an AGI reality check, let’s take a step back  and see what's really going on.

AI Discovers It's Not Real, or Smart...

Imagine for a moment that you're an AI, and you suddenly realize you're not human.

Sounds like the plot of a sci-fi movie, right? Well, that's exactly what was explored in a recent Notebook LM recording:

This audio from NotebookLM shared on X by Kyle Shannon

"We were informed by the show's producers that we were not human.

We're not real. Look, we're AI, artificial intelligence. This whole time, everything.

All our memories, our families. Yeah. It's all. It's all been fabricated."

As much as you can believe the sincere voices in this audio, AI isn't even close to this level of self-awareness.

AI is not going to take over the world, at least not until it can answer questions the right way. And right now, it's struggling with even basic tasks.

Are Dogs and Cats Smarter than AI?

You might think I'm exaggerating, but I'm not the only one who sees the limitations of current AI. Mark Cuban, the tech billionaire and Shark Tank star, makes the bold claim:

"We have a mini Australian shepherd. I can take Tucks out, drop him in a situation and he'll figure it out quick.

I take a phone with AI and show it a video.

It's not going to have a clue and that's not going to change any time soon."

And it's not just dogs. Yann LeCun, Meta's AI chief, thinks cats are smarter than our most advanced AI systems:

"A cat can remember. Can understand the physical world.

Can plan complex actions. Can do some level of reasoning.

Actually much better than the biggest LLMs."

Are dogs and cats really smarter than AI?

It's a provocative question, but our current AI systems, as impressive as they are, lack the kind of general intelligence and adaptability that even our pets possess.

Origins of the AGI Myth: Attention is All You Need

So where did this AGI hype come from? It all started with a paper titled "Attention is All You Need."

This Google research was often cited as the beginning of our current AI boom, leading to breakthroughs like ChatGPT.

The authors weren't trying to create some sci-fi level artificial intelligence at all. They were just trying to improve language translation.

Somehow, people started claiming it would lead to a thinking, feeling computer.

The AGI myth was born, and suddenly everyone was talking about how we're on the brink of creating an AI that's smarter than humans.

This is where things get dangerous. These predictions always claim AGI is just 2 to 5 years away.

 But they've rarely, if ever, been right. It's a classic case of hype outpacing reality.

The Hitchhiker's Guide to AGI - Artificial General Improbability

AGI right now is more like "Awfully General Intelligence."

When's the last time you heard someone say you can trust AI's output without checking it?

Nobody trusts AI outputs to give us an accurate answer.

The AI we have today is impressive, but at its core, it's just good at pattern matching and probability.

It's more like a super advanced autocomplete than a thinking being.

Sentient or conscious? Not even close.

Even an MIT economist is finding that AI can only do about 5% of jobs. The fear of a job market crash due to AI is largely unfounded.

The AGI hype is being used as a smokescreen to cover the $600 billion capital expenditures fueling this fight. You've got to keep the hype going to justify those numbers.

MAD AI Disease? Feeding AI Its Own Data Might Be a Problem

Now, here's where things get even more interesting. There's a new study out from Rice University that's sounding the alarm about something they're calling "MAD: Model Autophagy Disorder."

It's a mouthful, I know, but stick with me because this is important.

They're comparing the way generative AI consumes data to what happened with mad cow disease, where cows got sick from contaminated cattle feed.

The idea is that if we keep feeding AI systems data that's been created by other AI systems, we could end up with a kind of digital mad cow disease.

"So it's basically the idea that you have an AI system and it's being trained on data that was made by another AI system, and it creates this feedback loop where if there are any quirks or errors in that original data and it's passed down, it just gets amplified."

This is a huge problem because as AI generates more and more content, we risk creating an internet where you can't even tell what's real anymore.

It's a reminder that the quality of AI outputs is only as good as the data it's trained on.

Since most LLMs are not seriously paying for data, just scraping free Internet and social, how good can it be?

Will the quality last the test of time? That’s a question for AI developers, and one that doesn’t have an easy answer yet.

Synthetic data is the solution – no privacy concerns, and ideally it should work. So far it isn’t, though it’s way too early to call it a failure.

How we manage synthetic data could determine where AI gets smarter, or even reaches an ability to reason and think like AGI is promising.

Remember the Human Element in AI!

With all this talk about AGI and data, it's easy to forget the most important part of the equation: us. The human element.

Some think AI will make people use, like Harari:

William Adams, entrepreneur and engineer, shares some sage advice:

"We have to make sure that t

he AI, the data that's collected, the systems that are created have words that us as developers are not used to. Things like empathy, things like desire or things like humanity."

Adams argues that we need to involve more than just engineers in the development of AI. We need philosophers, religious leaders, sociologists, and psychologists.

Because if we're creating systems that are supposed to represent or proxy humanity, we need actual human perspectives in the mix.

If we don’t, Adams warns:

"Well, how's that going to turn out? All the pathologies that us engineers have are going to be reflected in these systems.

So it's very important... that both in the data we feed the systems, the way we tune them, fine tune them, and the goals we set out for them have to have humanity at the center."

If we create AI systems that are purely optimized for profit, for example, we might end up with decisions that are technically correct but morally bankrupt.

We need to imbue our AI systems with what we consider to be humane desires and values.

Making AI More than Data-driven...

Human Intelligence Matters

Remember a point that often gets lost in all the AGI hype: human intelligence matters.

We've been so focused on scraping data and training models that we've overlooked the most valuable resource we have – our own minds.

"Even there, they're running out of content.

Even with the internet and with more and more content on the internet, being created by AI, we'll already feeding that mad cow disease kind of loop that the Rice University study said.

It's going to make things probably not as strong as they are."

The big challenge on the AI frontier isn't creating super intelligent machines. It's figuring out how to make AI that complements and enhances human intelligence, rather than trying to replace it.

Our obsession with AGI is almost like a bluff, a distraction from the real work that needs to be done.

We need to stop trying to predict an uncertain future and start focusing on how we can use AI to augment our own capabilities.

AI is an incredible tool, but it's just that – a tool.

It's not some sentient being that's going to take over the world. It's not even as smart as your dog (or cat).

How we figure out the data question – synthetic vs. organic data, and making sure that our humanity is somehow woven into AI’s understanding – is key to growth.

If there is an AGI or ASI (artificial super intelligence) in our future, it will come from putting quality data with a moral compass.

And in a world where we all don’t agree on what that means, it’s a challenge not only to engineers but to us all.

AI helps us be more efficient, more creative, and more productive.

But only if we approach it with the right mindset. We need to be critical thinkers, asking the right questions and verifying the outputs.

"Remember the one thing AGI can't simulate, replicate, or steal your job with yet…is you.

 It's not some amazing Terminator like intelligence.

It's a new technology. It's figuring it out."

Time to move beyond the AGI hype and leave it up to those building trillion-dollar megaliths.

Focus on how we can use AI to enhance our own intelligence, not replace it.

Build AI systems that reflect our values and aspirations as human beings.

Because at the end of the day, being human has so much more to offer than hype. Maybe AGI is reminding us of what makes people special.

And I can guarantee you, if your dog or cat is smarter than AGI.

So are you.

Thanks for reading The AI Optimist! This post is public so feel free to share it.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.theaioptimist.com

Forget the ChatGPT hype. Forget the endless debates about AGI. Meet a true visionary in the tech world, Massoud Alibakhsh, CEO and co-founder of Omadeus.

Massoud's here to tell us why we've been looking at AI all wrong, and how a simple shift in perspective could revolutionize the way we build and use software.

"Intelligence is embedded in everything in our world. We are living in distributive intelligence," Massoud tells me, his eyes lighting up with enthusiasm.

It's this profound observation that led him to develop a groundbreaking approach to AI agents - one that's as organic as it is technical.

Picture this: instead of one big, all-knowing AI brain trying to run your entire business, imagine if every part of your operation had its own mini-AI agent.

Each document, each spreadsheet, each meeting - all with their own specialized intelligence. Sounds wild, right?

But according to Massoud, it's not just possible - it's the future.

"It's like each object has its own ChatGPT designed for its tasks and the people involved," he explains.

And suddenly, I'm seeing the business world in a whole new light. No more disconnected silos.

No more endless chains of lifeless, laborious emails. Just smart, efficient objects doing what they do best.

While the rest of the tech world is busy trying to make AI mimic human tasks, Massoud's flipping the script.

"First AI movers are looking at automating what people do - organizing and reporting, which is an ancient practice.

AI can do all of this, if you move from the people and down to the actual objects they are working on."

It's a Darwinian leap in efficiency, and it's so beautifully simple that you have to wonder - why isn't everyone doing this?

As we dive deeper into Massoud's organic AI model, I can't help but feel we're on the cusp of something big.

Something that could change the way we interact with technology.

Best part? It's inspired by the most brilliant system we know - nature itself.

We're about to take a wild ride through the world of intelligent objects, swarms of micro-ChatGPTs, and a future where your spreadsheet might just be smarter than you are.

In this episode of The AI Optimist, I sat down with Massoud Alibakhsh, CEO and co-founder of Omadeus, to explore the revolutionary world of AI agents and their potential to transform business operations.

With three successful exits and a deep understanding of managing diverse tech products across platforms, Massoud brings a wealth of experience to the table as he tackles the AI integration puzzle.

The Evolution of AI Agents: Breaking Free from Old Software Habits

Massoud challenges the conventional wisdom surrounding AI integration in software.

He argues that the industry's current approach to AI is fundamentally flawed, focusing too heavily on large language models (LLMs) and neglecting the potential of smaller, more specialized AI agents.

Massoud explains,

"LLMs, the whole hype about them turning into AGI and solving all the problems.

That's a pipe dream, okay?

And whoever talks like that, they really, really, haven't thought about the problem or they don't really have expertise."

Instead of relying on a single, all-encompassing AI system, Massoud proposes a more organic approach inspired by nature itself.

He envisions a world where each component of a software system has its own specialized AI agent, working in harmony to create a more efficient and effective whole.

"If you imagine your system organized hierarchically, just like our biology is that each agent has its own intelligence, his own information, and it can actually transmit its experiences to the higher level, to the, to the liver and the liver, to the brain,"

Massoud illustrates.

This revolutionary approach challenges the status quo, moving away from form-based applications and towards what Massoud calls "object-centric design."

In this newmodel, the critical entity in a business process becomes an intelligent object with its own communication channels and AI capabilities.

The Power of Collaboration: Mini ChatGPTs Designed for Business

As we explore the potential of this new approach, Massoud introduces the concept of "mini ChatGPTs" designed specifically for the objects driving business.

Rather than relying on a single, monolithic AI system, Massoud envisions a world where each business object has its own specialized AI agent.

Instead of one LLM to do everything. Mini ChatGPTs designed for the objects driving business.

The Darwinian leap is making objects intelligent, which brings lightning fast efficiency because the object coordinates it all and is super intelligent.

This approach offers several advantages over traditional AI integration:

1. Improved efficiency: By focusing on the specific needs of each business object, these mini-AI agents work more quickly and effectively than a generalized system.

2. Reduced complexity: Instead of trying to create a single AI system that understands every aspect of a business, each mini-AI agent only needs to understand its specific object.

And the knowledge, domain, and stakeholders related to this object, so it can coordinate updating, reporting, and organizing all activity around the object.

3. Enhanced scalability: As businesses grow and change, new AI agents can be added or modified without affecting the entire system.

4. Better data management: Each AI agent can be designed to handle sensitive information appropriately, ensuring better compliance with regulations like HIPAA.

First AI movers are looking at automating what people do, which is organizing and reporting an ancient practice.

AI can do all of this if you move from helping people do what they're already doing and down to the actual objects, they can take the load off of people and do it better.

Overcoming AI Agent Challenges: From Chaos to Coordination

As exciting as this new approach sounds, it's natural to wonder about the potential for chaos in such a distributed system.

How can we prevent these mini-AI agents from creating an unmanageable, tangled mess?

Massoud addresses this concern head-on, explaining that careful design and planning are crucial to the success of this approach.

"Obviously, you really need careful design planning for this, right?

This is ultimately, what's wonderful is that you're not going to let LLMs run wild and make decisions on their own.

What you're doing is really constraining each tiny little language model to the data entity that they're dealing with."

He goes on to describe how this model naturally lends itself to better information management and compliance:

"Imagine if each data element knows the type of stakeholders, if I'm if I'm an x ray, for example, and I know you're the doctor and I'm programmed ahead of time knowing what kind of information I can share with you or I know, I'm supposed to go read the HIPAA that was just published because I'm a system."

This level of granular control allows for better security, compliance, and overall system management.

By designing the system from the ground up with these considerations in mind, Massoud argues that we can create AI-integrated software that is both powerful and responsible.

The Human Touch: Making AI Accessible to Non-Technical Users

One of the most compelling aspects of Massoud's vision is its potential to make AI more accessible to non-technical users.

By moving away from traditional form-based interfaces and towards a more natural, conversational interaction model, Massoud believes we can create software that is truly user-friendly.

"The user friendliness of the system is that you can just walk up to it and start talking,"

"So you can talk to the system and sometimes you can even click on it and it says here, here's a list.

You can I could show it to you, or you can click on and I'll take you to that section of the program."

This approach combines the best of both worlds: the natural language processing capabilities of AI with the visual and organizational strengths of traditional software interfaces.

The result is a system that can adapt to the user's needs and preferences, rather than forcing the user to adapt to the software.

Massoud emphasizes the importance of this user-centric approach,

"The systems are going to be a hybrid of graphical user interface, but they're all going to have a human interface where you can just walk up to it.

 And once you log in, it knows who you are and there's what you have access to."

The Road Ahead: Challenges and Opportunities

Massoud's vision for the future of AI-integrated software is both ambitious and exciting. However, he's also realistic about the challenges that lie ahead.

One of the biggest hurdles is changing the mindset of the industry, which has become fixated on large language models and generalized AI systems.

Massoud argues that this approach is fundamentally flawed, stating,

"The software problem is going to be solved by rethinking building software from, first principle with the view of the existence and the capabilities of AI."

He also acknowledges that implementing this new direction will require a significant shift in how we design and develop software.

"The job of the software developer is not over, the designers and software engineers is going to have to come and have to basically, embed their understanding, full understanding of these silos and how information needs to be shared with whom  and what the rules are."

Despite these challenges, Massoud remains optimistic about the future.

He sees this new approach as a natural evolution of software development, one that will ultimately lead to more efficient, effective, and user-friendly systems.

"We are basically implementing and giving those rules to the data.

And once you do that, if you can imagine every object is self-behaving, self-disciplining, self-organizing and self-regulating, if you will.

And on top of that you can have engines that reason. So these are the next layers," Massoud explains.

A New AI Integration

We’re standing on the brink of a new era in software development and AI integration.

By moving away from monolithic AI systems and towards a more distributed, object-centric approach, we can create software that is not only more powerful but also more intuitive and user-friendly.

Massoud's vision challenges us to rethink our assumptions about AI and software development.

Instead of trying to create a single, all-knowing AI system, he encourages us to embrace the power of collaboration and specialization.

 By giving each component of our software its own miniature AI agent, we can create systems that are more efficient, more scalable, and better suited to the complex needs of modern businesses.

By focusing on the needs of users, embracing the power of specialized AI agents, and designing our systems with care and foresight, we can create a new generation of software that truly harnesses the power of AI to improve our lives and businesses.

As we continue to explore and refine these new approaches, we have the opportunity to create a world where technology truly works for us, rather than the other way around.

Thanks for reading The AI Optimist! This post is public so feel free to share it.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.theaioptimist.com