AI Daily

In today's episode of AI Daily, we bring you the latest news from the world of AI. First up, we discuss the Google AI Pact with the EU. As Europe pushes for AI regulations, Google is the first corporate partner to join this initiative. We delve into the importance of this partnership and the potential impact it may have on AI regulation worldwide. Next, we explore the exciting Alexandria Initiative, which aims to embed the entire internet. This project, led by microcosm, focuses on making research papers more accessible and improving collaboration and search capabilities. We discuss the benefits and potential applications of this groundbreaking technology. Lastly, we dive into the soaring growth of NVIDIA and the buzzing discussions around its stock. While NVIDIA's market cap continues to rise, we caution against FOMO (Fear of Missing Out) and remind viewers to consider other players in the market. Join us for this informative episode of AI Daily!

Key Points

Google AI Pact with EU:

* Google and the EU are in discussions similar to OpenAI and the US government.

* Europe is pushing for AI regulation, aiming to implement it by the end of the year.

* Google is the first corporate partner to join the AI pact with the EU.

* There are concerns about the lack of public involvement and actual actions in AI regulation discussions, with accusations of virtue signaling by politicians and companies.

Alexandria Initiative to Embed the Internet:

* Microcosm has developed Alexandria, a project aiming to embed the entire internet, starting with titles and abstracts of research papers from archives.

* Alexandria improves search and collaboration by providing easier access to research papers and enhancing semantic web capabilities.

* The project is open source, allowing users to download the models and participate in the open collective to decide what content to embed next.

* The advancements in embedding technology, such as cheaper and more useful embeddings, along with open source models, have made projects like Alexandria possible now.

NVIDIA’s Growth:

* Nvidia's stock market value is skyrocketing, prompting discussions on whether it's still a good time to buy the stock.

* While the consensus is that Nvidia will continue to grow, there is a cautionary note about other players entering the market and potential price fluctuations.

* The situation is reminiscent of internet-driven hype and market volatility, where stocks can experience significant increases and decreases.

* Nvidia is seen as having a monopoly on GPUs and AI infrastructure, but the emergence of custom boards and other companies may challenge their dominance in the long term.

Links Mentioned:

* Google-EU AI Pact

* Alexandria Embed

* Nvidia Growth

* Diagram

* UI AI

Follow us on Twitter:

* AI Daily

* Farb

* Ethan

* Conner

Subscribe to our Substack:

* Subscribe

Join us for another exciting episode of AI Daily as Conner, Ethan, and Farb discuss the latest breakthroughs in artificial intelligence. In our first story, we explore the remarkable performance of the Goat LLaMA model, which surpasses GPT4 and other models in arithmetic tasks. Discover how synthetic data and fine-tuning techniques contribute to its success. Then, we dive into Adobe Photoshop's new tool, Firefly, designed to revolutionize photo editing. Learn how this powerful tool can enhance productivity and transform design workflows. Finally, we explore the AlpacaFarm Simulation Framework, a game-changing approach to reinforcement learning using synthetic data and feedback. Find out how this framework can revolutionize the training of AI models. Don't miss this episode filled with exciting advancements in AI technology.

Key Take-Aways:

Goat LLaMA Model

* The Goat LLaMA model, a fine-tuned 7 billion parameter model, outperforms GPT4 and Palm 540B on arithmetic tasks.

* The use of synthetic data in training the model is an interesting approach to improve performance.

* Fine-tuning a model specifically for arithmetic tasks raises questions about the nature of learning mental math and the relationship between language and mathematics.

* The methodology of taking a small model and fine-tuning it to outperform GPT4 on a challenging task is significant and holds potential for future tasks.

Adobe Firefly

* Firefly, a new tool for Photoshop, has been released and is expected to greatly enhance productivity for designers and photographers.

* The tool, which was trained on Adobe Stock, offers a safe and reliable option for businesses to utilize generative AI technology without copyright concerns.

* Firefly's integration with Photoshop is just the beginning, as similar advancements are expected in video and audio editing.

* While there are other tools available for generative AI, Firefly and Adobe provide a trusted solution for professionals and offer unique copyright advantages.

AlpacaFarm Simulation Framework

* Alpaca Farm Simulate is a simulation framework for training methods in reinforcement learning through synthetic data and feedback.

* The framework offers a cost-effective and efficient way to train models for chat and fine-tune them without relying solely on human feedback.

* Synthetic data and simulations can save significant time and resources compared to using real human data, making it a practical approach for training models.

* Alpaca Farm Simulate demonstrates the potential of simulations and synthetic data in improving models, offering an alternative to expensive and time-consuming human-centric approaches in reinforcement learning.

Links to Stories Mentioned:

* Goat LLaMA Model

* Adobe Firefly

* AlpacaFarm Simulation Framework

* Google Merchant Center

* Microsoft Build

Follow us on Twitter:

* AI Daily

* Farb

* Ethan

* Conner

Subscribe to our Substack:

Subscribe

In today's episode of AI Daily, we bring you three exciting news stories that will leave you wanting to know more. First, we dive into the world of Microsoft with their latest announcement, Windows Copilot, a game-changing addition to the Windows platform. Next, we explore Meta's groundbreaking language model that is revolutionizing speech-to-text and text-to-speech capabilities. Discover how this multilingual model is outperforming its competitors and supporting over 1,100 languages. Lastly, we delve into the thought-provoking topic of governing superintelligence, as OpenAI sheds light on the potential risks and solutions. Join us as we unravel the complexities of AI governance. Don't miss out on these fascinating stories - tune in now!

Main Take-Aways:

Microsoft Windows Copilot

* Microsoft made several announcements, including two main ones: Windows Co-pilot and plugins across Bing and ChatGPT.

* Windows Co-pilot is being brought to all of Windows, offering integration on the right side of the Windows home screen and allowing users to ask questions, drag in files, inquire about apps and system settings.

* Despite Windows' support for older versions like Excel 95, Windows 11 and Windows Co-pilot are praised for their integration and functionality.

* Microsoft claims to have more AI-capable GPUs on Windows 11 than any other operating system, emphasizing the vast scale of devices that can leverage these technologies.

* Windows Co-pilot is set to become available in June, making it a significant move for Microsoft and potentially influencing developers' preferences and workflow choices. The system-level integration is highly anticipated, and it will be interesting to see how Mac and Apple respond in the coming years.

Meta Language Model

* Meta has introduced an absolute open-source model focused on speech-to-text and text-to-speech capabilities, outperforming Whisper and supporting over 1,100 languages.

* The multilingual model has significantly lower error rates compared to Whisper, with model sizes half as large, making it a powerful tool for various tasks.

* The model's training data includes religious texts, leveraging their widespread availability and making it suitable for many languages.

* Despite Meta's minimal press coverage, their advancements in AI technology rival those of major players like Microsoft, OpenAI, and Google.

* With over 7,000 languages globally, Meta's model covers approximately 1,100 languages, some of which are at risk of disappearing, providing a valuable resource for preserving linguistic diversity. The model is entirely open-source, accessible on GitHub for developers and researchers.

Governance of Superintelligence

* OpenAI recently released a blog post discussing the governance of superintelligence and the potential risks associated with AI becoming expert-level in multiple domains within the next decade.

* They proposed various approaches to address these risks, including limitations on GPU access and model training for large-scale AI models, as well as the establishment of government agencies to oversee regulation.

* An interesting point raised was the suggestion that regulation should only apply above a certain capability threshold, leaving lower-level AI systems ungoverned.

* OpenAI emphasizes their intent to explore and experiment with plausible solutions openly, seeking input from the global community rather than claiming to have all the answers.

* The comparison to nuclear energy and the reference to the International Atomic Energy Agency (IAEA) highlight the need for careful consideration and potential governance measures as AI advances.

Links to Stories Mentioned:

* Microsoft Windows Copilot

* Meta Language Model

* Governance of Superintelligence

* Instruction-finetune(IFT) Stable Diffusion(SD)

* Anthropic Series C Funding

Follow us on Twitter:

* AI Daily

* Farb

* Ethan

* Conner

In this episode, we delve into three exciting news stories. First up, we explore the remarkable LIMA model, a 65 billion parameter model that performs almost as well as GPT4 and even outperforms Bard and Da Vinci! Find out how Meta's innovative approach is revolutionizing the open source AI realm. Next, we unravel the fascinating CoDi model, which utilizes compositional diffusion for generating multimodal outputs. Learn how this powerful model can transform text and audio inputs into stunning videos. Lastly, we uncover the mind-boggling Mind-to-Video technology that reconstructs videos based on brain activity. Join us as we discuss the possibilities and implications of mapping the human mind. We also discuss hilarious ChatGPT App reviews, which you won’t want to miss!

Main Take-Aways:

LIMA Model

* The Lima model is a large 65 billion parameter LLaMa model that was fine-tuned on a thousand carefully curated responses.

* It performed almost as well as GPT4 and even better than models like Bard and Da Vinci.

* Meta has released the Lima model as an open source model, showcasing their commitment to staying up-to-date in the AI field.

* Lima's approach differs from other models by using supervised examples instead of human feedback, resulting in impressive responses.

* This development is significant because it brings another open source model closer to competing with the massive models trained by OpenAI, providing an alternative approach to alignment.

CoDi

* CoDi is a model that specializes in any-to-any generation using compositional diffusion.

* Unlike other models that primarily process text, CoDi is designed for multimodal tasks, allowing users to input combinations of text, audio, and video to generate corresponding outputs.

* CoDi can generate videos based on text and audio inputs or produce new text outputs based on two different text inputs.

* Understanding context is crucial for CoDi, as it needs to comprehend the relationships between different modalities to provide accurate and comprehensive results.

* CoDi appears to be an open-source model, potentially an enhanced version of Meta's previously released ImageBART, although a direct comparison has not been made yet. The code for CoDi is likely available for use.

Mind-to-Video:

* The team has developed a mind-meets-video model that reconstructs videos based on brain activity captured through fMRIs.

* The training process involves pairing fMRI data with corresponding videos, allowing the model to learn the relationship between brain signals and video content.

* The model aims to capture what a person is remembering or perceiving by analyzing their brain activity and finding similar videos from the training set.

* Although it is not yet capable of mind reading, the model provides insights into how the brain processes and represents visual information.

* The team's previous work focused on mind-to-image generation, and this mind-to-video model represents an impressive advancement, achieving a 45% increase in accuracy compared to previous methods.

Links to Stories Mentioned:

* LIMA Model

* CoDi

* Mind-to-Video

* Lambda Demos

Follow us on Twitter:

* AI Daily

* Farb

* Ethan

* Conner

Today on AI Daily, we kick things off with Blockade Labs' mind-blowing 3D Skybox generation tool that turns your sketches into stunning scenes. Watch as Farb demonstrates the power of this incredible tool. Then, we dive into Drag Your GAN—a game-changer for image manipulation. Discover how you can effortlessly transform images with a few simple clicks. But that's not all! Join us as we explore Meta's latest innovation, the MTIA v1 AI inference accelerator. Find out how this custom ASIC board takes AI operations to unprecedented heights. Don't miss out on this epic episode filled with revolutionary technologies and endless creative possibilities.

Main Take-Aways:

3D Skybox Generation

* Blockade Labs has developed a free 3D Skybox generation tool.

* Users can sketch out their desired skybox and provide a prompt.

* The tool generates an entire scene based on the sketch and prompt.

* It offers a quick and easy way to generate ideas and scenes.

* The generated skybox can be used as a background in games or other forms of art.

* The tool is accessible to anyone and can be used immediately.

* It is speculated that stable diffusion and control net algorithms are used in the process.

* The tool has the potential to impact 3D game developers and artists.

* It simplifies the creative process by combining multiple steps into a single tool.

* Users can feed the generated images into other models to create 3D assets.

* The current main use case is creating skyboxes for game backgrounds.

Drag Your GAN

* "Drag Your GAN" is a new GAN that allows interactive point-based manipulation of images.

* Users can drag points and adjust dials to manipulate specific features in the image.

* The tool enables realistic and seamless transformations of objects or faces in images.

* It offers fine-tuned control over image poses and allows for easy adjustments without extensive editing.

* The tool has examples demonstrating its effectiveness in manipulating cars, microscope images, faces, and animals.

* It simplifies the process of fine-tuning image outputs and provides controllability beyond the base-level image models.

* The ability to create masks over specific areas of an image allows for selective manipulation while keeping the rest of the image intact.

* The tool is expected to be released in June, and creators are planning to share the code repository on GitHub.

* It offers a more efficient and user-friendly alternative to image editing in software like Photoshop.

* The tool has the potential to significantly benefit creatives and make image manipulation easier and more accessible.

Meta MTIA v1 Inference Accelerator

* Meta has announced their MTIA v1, the first generation AI inference accelerator.

* It is a custom ASIC board designed specifically for AI operations, focusing on specific AI math requirements.

* The accelerator is different from Google's TPU and shows promise in comparison to other hardware options like etch.

* It excels in handling small shapes and batch sizes, but GPUs are still more efficient for medium or large size shapes.

* The integration of the accelerator with PyTorch is a significant advantage, providing backward compatibility and facilitating inference workflows.

* Energy efficiency is a key feature, with the ability to run a single accelerator using only 35 watts of power.

* The development of the accelerator has been in progress since 2020, indicating extensive effort and refinement.

* Meta's release of this custom hardware demonstrates the growing competition in the chip race, with companies like Apple, Google, and Microsoft also investing in custom hardware solutions.

* The integration of hardware accelerators like MTIA v1 has the potential to improve inference performance and reduce energy costs for a range of applications.

* The future looks promising for further advancements in custom hardware for AI.

Links to Stories Mentioned:

* Blockade Labs 3D Skybox Generation

* Drag Your GAN

* Meta MTIA Inference Accelerator

* Perplexity Copilot

Follow us on Twitter:

* AI Daily

* Farb

* Ethan

* Conner

First up, OpenAI has released the official ChatGPT mobile app, bringing the power of chat GPT right to your fingertips. Discover its smooth interface, voice recording, and haptic features that make it a joy to use. Next, we explore the integration of AI coding assistant in Google Colab, making it easier than ever to learn and experiment with AI models. And last but not least, we delve into the advancements of MedPaLM2, Google's game-changing healthcare model. Discover its impressive accuracy and potential impact on the healthcare industry.

Main Take-Aways:

ChatGPT Releases App

* OpenAI has released an official ChatGPT mobile app.

* The app is smooth, clean, and has nice features such as haptic feedback, voice recording, and voice memos.

* It is faster than native iOS voice transcription and allows users to search their history.

* The app does not have browsing or plugins yet, but those features may be added in the future.

* It uses GPT-4 and is snappy, providing quicker access to chat.

* An Android version of the app is expected to be released soon.

* Some users have already started using the app and replacing Safari and Chrome on their iPhones.

* There are many fake Chat GPT apps in the app store, but OpenAI's official app should help eliminate them.

Google Colab

* Google's CoLab, a Jupiter notebook environment, is integrating an AI coding assistant.

* This integration aims to make coding and AI learning more accessible to users.

* The AI coding assistant in CoLab is built off of Palm II, a language model developed by Google.

* The assistant offers autocomplete suggestions, a chatbot, and a generate feature to create new code blocks.

* Users can benefit from AI assistance within CoLab, enhancing their coding experience.

* The integration is seen as a significant improvement for CoLab, attracting users back to the platform.

* It is expected to have a positive impact on the developer community, particularly for beginners learning to code.

* The AI coding assistant is not yet available but will be rolled out to paid subscribers first and eventually to all users.

* The code blocks feature is free, while autocomplete and the chatbot are available to paid users.

MedPaLM2 for Healthcare

* MedPaLM2, Google's healthcare-specific model, has shown improvements in accuracy.

* MedPaLM2 achieved a score of 86.5% on the Med QA exam and received positive evaluations from physicians and patients.

* The model provides quick responses and aims to assist clinicians and potentially patients directly in the future.

* The integration of AI models like MedPalm 2 into healthcare raises legal and ethical considerations, such as liability for decisions made contrary to AI recommendations.

* Technical details of MedPaLM2's advancements include adversarial question testing and efforts to increase accuracy to 100%.

* The development of application-specific integrated circuits (ASICs) for running language models (LLMs) offers significant performance improvements compared to GPUs.

* Microsoft and Apple are also investing in AI-specific chips, indicating the growing trend in the industry.

Links to Stories Mentioned:

* ChatGPT App

* Google Colab

* MedPaLM2

* Etched

Follow us on Twitter:

* AI Daily

* Farb

* Ethan

* Conner

Welcome to AI Daily! In today's episode, we bring you three exciting news stories that you don't want to miss. First up, Stability AI introduces Stable Studio, an open-source version of Dream Studio that allows easy integration with their latest models. Discover how this web application is revolutionizing the way people use stable diffusion models. Our second story dives into an intriguing incident at Texas A&M, where a professor accused students of using ChatGPT in their writing, causing many to not only to fail the class, but possibly jeopardize their degree all together. Find out why this story is causing a stir and what it means for the future of AI in education. Lastly, Microsoft unveils Guidance, a powerful templating language that simplifies working with LLMs. Learn how this framework saves time and empowers developers to create cleaner, more efficient applications. Join us for an engaging discussion on these trending topics and stay ahead of the AI-curve!

Main Take-Aways:

Stable Studio by Stability AI:

* Stability AI announces Stable Studio, an open-source version of Dream Studio.

* Stable Studio is a web application that allows users to work with stable diffusion models easily.

* The purpose of Stable Studio is to provide a platform for users to keep up with the latest models, integrate custom models, and extend the capabilities of stable diffusion.

* The release of Stable Studio emphasizes StabilityAI's commitment to open source and their goal of making it easy for people to use their models.

Texas A&M Professor failing students for using ChatGPT:

* A professor at Texas A&M University wrongly accused students of using ChatGPT to write their essays.

* The professor fed the essays into ChatGPT, which confirmed their involvement, leading to the students' failure and graduation blockage.

* The incident highlights the misconception that ChatGPT keeps a log of everything it generates, leading some to blindly trust its statements.

* The lack of clear policies and guidance from educational institutions on AI use and assessment raises concerns and the need for proper communication and understanding.

Microsoft Guidance:

* Microsoft launches "Guidance," a templating language designed to work with LLMs (large language models).

* Guidance simplifies the process of working with LLMs by allowing developers to specify desired outputs and template variables.

* It supports various LLMs, such as GPT-4 and Vaya, providing flexibility in model selection.

* The introduction of Guidance aims to address challenges in controlling LLM outputs and enables developers to create cleaner and more efficient applications.

Links to Stories Mentioned:

* Stable Studio

* Texas A&M Professor Story

* Update

* Microsoft Guidance

Follow us on Twitter:

* AI Daily

* Farb

* Ethan

* Conner

Transcript

Farb: Good morning and welcome to AI Daily. I'm Farb Nevi. I'm here with my excellent, uh, co-hosts and esteemed team members, Ethan and Conner, and we have three great pieces of news for you today. Let's jump into it. The first one is a big piece of open source news from the folks at Stability they've announced Stable Studio, their open source version of Dream Studio. Can you tell us about this, Ethan?

Ethan: Yeah, so Stable Studio is just like you said. So Dream Studio has been stability, AI's way to use stable diffusion online. So a really easy portal to come in play with stable diffusion 1.5, stable diffusion. Two, their new language model, stable lm. And they've dropped Stable Studio as really an open source version of this kind of web application.

They really want something that people can continue to keep up to date with their latest models, plug and play, maybe custom models. They've been fine tuning on stable diffusion models and trying to make something that people can integrate with better. Utilize better, kind of extend upon and just continue to place themselves in the forefront of open source.

If you've played with these models and you're using 'em open source, maybe you're using something like radio or maybe you're using, you know, one of those tools like that online and Stable Studio is really their answer to this. They want to continue to embed themselves in open source and make it easy for people to use their latest models.

Farb: Why is this important, Connor, and what could you do with it?

Conner: I think the, one of the most important things here is that they announce plug-ins. So instead of directly calling stability APIs, now they have plug-ins built into stable studio. I think their angle here is try to give something to the community to let the community build upon and hopefully even improve Dream Studio itself in the future, I would imagine.

Mm-hmm. Um, They probably see no benefit really in keeping Dream Studio closed sourced. There's many other alternatives. So now the benefit for them is to make it open source, stable studio. If you have a good plugin idea, go build a

Farb: plugin in for developer. So, yep. Is that the main thing you think is, you know, one would do with it, uh, is, is build plugins for it.

Conner: I imagine. I mean, there's, like Ethan said, you can use radio right now, you can use. Um, automatic one 11 web UI to generate other content right now. So if you wanna build plugins for this, build plugins for

Farb: this. It seems like the, the sort of make sure the open source community is getting love from your organization.

Your AI project, uh, continues to be very important if you. Sort of stop doing that. I think people will start just recognizing you as some sort of, you know, over, over there in the corner team that's trying to keep everything for themselves. And you're not gonna get the, uh, the masses, you know, teaming up around what you're doing and sharing what you're doing, and you might just sort of slowly disappear.

Uh, if that sort of stuff is important for you. Some companies are big enough that, you know, even the big guys like Google obviously care a lot about this. Uh, the folks at Apple you see aren't doing necessarily as much right now with the open source world, but um, you know, they're obviously big proponents of the of, of Dev, so maybe they don't really need to do that as much, uh, as some of the other folks do to show, show the dev world their love.

Very cool, uh, to see more awesome open source stuff happening. Let's go on to our second story. Uh, this is a little bit more of a social story. The, um, you know, professor at Texas A&M, it seems like, has maybe incorrectly labeled a bunch of folks as using chat G p T to write essays, even though that's not what happened.

Connor, could you tell us a little bit more about this and why is this story important? Yeah,

Conner: apparently the story is that he got his entire graduating class and he fed their essays in Cheche, BT and Chei. Bts like, yes, I wrote that and based off that, he failed all them and is blocking their graduation.

Uh, it's a pretty common thought process among people who use Chei b t. They think Cheche BT keeps a log of everything. It says the huge, they think chat. BT knows what it says, and they asked Shay Bt if it wrote something, and Shay Bt says, yes. And then they just believe it.

Farb: So yeah, chat,GPT sort of, you know, probably wants to validate what they're saying to it and it, it, you know, it's like, if you did this, it's maybe it just wants to generally apply in, you know, the affirmative, uh, it, it's interesting, it seems weird to me that.

You know, more schools aren't getting ahead of this type of stuff, and maybe they are, maybe they're quietly doing things and they're not ready to come up with announcements. But to, you know, this isn't the first time we've heard this type of story. Uh, this has been happening since, you know, last, last fall, when a lot of these, you know, cooler, more powerful, LLMs started, started hitting the, the public.

Uh, this has been going on. It seems a little strange to me that. You know, school the size of Texas a and m doesn't, hasn't already announced some sort of policy or provided its professors with some guidance on how to do things. E Ethan, where do you think this is going? Is this the end of this? Is this just the beginning?

Ethan: No, I, I think you're on point with that last point. I was just talking to an educator friend of mine yesterday, and at the end of the day, I, I get it from a professor's point of view, from a teacher's point of view, you're like, Hey, why am I grading this stuff when they're just taking five minutes to write it with chatGPT?

But from an administration side, from the top down level of these institutions, it needs to be made clear that. You chat, g p t. You can't just tell if this was AI written yet. Uh, G P T zero and chat, g p t saying, yes, I wrote it. None of these are accurate yet. So it's important that that is conveyed to these professors and that this is something we continuously deal with versus making these.

Like harsh reactionary points of saying, I'm just gonna fail everyone. I'm, I'm tired of this. I'm, if one person did it, maybe the whole class did it. I just wanna fail everyone. I'm sick and tired of grading ai, so there's some valid points, but it's, it's important that we're not doing things like this. So I feel bad for all these students who are sitting there who probably did write a good chunk of these thesis papers or their final papers and are now waiting for their degree because some AI models said,

Conner: yeah, that was I wrote that. Yeah, a few were already exonerated because, uh, because like they had like Googled docs like history of like edits and then like their emails and stuff were ignored by the dean and the teacher apparently. But as soon as these stories hit the Washington Post and everything got into the limelight.

Farb: Of course they will exonerate. It sounds like the definition of wishful thinking. You know, you, you want to be able to just plug these into Chad g p t and have it tell you if it's, if it's real or fake. And so you do it cause you're like, oh, if this thing's so powerful, it should be able to do this, I guess.

Uh, but it also seems like, you know, maybe this guy's, uh, just sick and tired of, of, of, of everything, you know? Um, yeah. I'm sort of sick and tired of this story, uh, but I don't think it'll, I don't think it'll be the last time we hear this story, quite frankly, maybe the last time that we talked too much about it.

So it'll have to become a bigger story. An entire, entire, you know, graduating classes of tens of thousands of people denied their diplomas or stuff for us to cover this again. But, you know, be careful out there folks. It's, uh, It's an AI world. All right, let's move on to our third piece of news. It's not small.

Um, you know, big news from, uh, stability announcing a, a new open source, uh, program. Um, and pretty huge news from Microsoft. Uh, launching what they call guidance, I think is they're just calling it guidance. Uh, and, and, and it's a language to help work with LLMs. Connor, can you tell us more about what exactly is this and, and why you think it's important?

Conner: Yeah, it’s a templating language, essentially like handlebars, so it's very easily, you can say exactly what you want and then you can template variables. But then more importantly here to integrate with LLMs. You can template generating for M, you can template. I want this part to have this temperature. I want this part to have this temperature.

I want this part to have this maxed tokens and you can use different lms. You can use anything from G P D four to Vaya. And Microsoft's writing something like guidance in such an open way that's such easy, so easy to template and can use any type of model, makes it really nice to see for Microsoft as a really like proponent of open source here.

Farb: So is this, is this really about, you know, doing much bigger things or is it more about saving time with the things you were already doing or some combination of both?

Conner: Both. I mean, you can, it saves a lot of time in these problems with templating we've seen before, like we talked about JSON former before.

About how difficult it is to get correctly formatted j s o out of a language model. So this does that, but also this makes it cheaper because, or not cheaper, but faster at least. Because instead of generating each part of a completion at all at once, it can separate it out. It can safely keep the tokens you need to be safe and it's another step forward in.

Farb: Guidance. Pretty much. Yeah. What, Ethan, what do you imagine, you know, assuming this is mostly for somebody who's developing, you know, and writing software on, on a regular basis, what do you imagine a, an engineer is doing with something like this? Yeah, I,

Ethan: I think at the end of the day, This is just seeing the space mature more.

If you've played with these language models a lot, especially on a code side or an application side, there, it becomes spaghetti code very quickly, uh, becomes an absolute mess very quickly trying to wrangle, okay, this prompt is here and then it changes to this prompt, but then I want it to do this. And when you're just trying to control the output of a language model, at the end of the day, Anyone been in the weeds of that?

You're looking at, you know, a bunch of functions and like I said, spaghetti code going everywhere. So Microsoft releasing a framework like this, we're just seeing the space mature. A few other people have done it, and I think Microsoft laid it out really well, um, with this guidance framework and just how do we control these models better?

How do we make applications better and how do we continue to make the feature set even larger? You know, cleaner code and cleaner applications give us. Bigger and better applications.

Farb: So excited to see, you imagine yourself using it?

Ethan: Oh, absolutely. Yeah. I think it's gonna be super useful, like I said, to remove all these spaghetti codes that people have been using before.

And you know, devs have put together their own mini frameworks to try to accomplish this, but having a. You know, it's almost like react to the web development world. It's a framework that everyone circles around, continues to add features to, is your base level framework is kind of what we're beginning to see these L LMS get.

Um, so I would compare this Microsoft guidance to, you know, some of the early web dev frameworks.

Farb: They really dialed in their GitHub page for it too. They weren't messing around. What were you saying Connor?

Conner: The current solution of this has been like link chain. As we, as we've seen, they've even raised a bunch of money to build this.

Um, but laingchain's got a lot of flack because it's over time, it's like Ethan said, gotten very spaghetti. They, it's kind of exploded in complexity because it wasn't structured for what it is now when it was originally built because these chat models didn't exist because the idea of agents didn't exist.

So doing all that in now is kinda hard, but with guidance agents is built in, chat models is built in. Far, but

Ethan: we were talking about Ruby on Rails the other day. You know, some of these web dev frameworks when you centralize them and you make them easy to use, you just enable more application developers, bigger projects, cooler features, and it's same thing's happening on the l om space right now.

So you're not manually doing all these prompts and chains,

Conner: et cetera.

Farb: Connor, do you imagine this replacing Laingchain in your, in your workflow?

Conner: It does for a lot of what it does. I would definitely see replacing link chain. Link chain still has some features like parsing files that this doesn't have built in yet.

But for agents and for chat and especially something like parsing json, this is definitely what replace Link Chain

Farb: my work on. Yeah, it's all json all the time around here. The, uh, is this available now? Yep. Do you know Open source now?

Conner: You can use all of it can contribute to it. It's here.

Farb: It's very, it's very cool.

What about the, the stable studio? I think that's available now too, right? Yeah. Yeah. Awesome. Yeah. All right. Couple of. You know, big stories plus a, uh, you know, some mistakes and errors going on over there at Texas a and m. Hopefully they can get their act together, um, and not embarrass themselves any further.

Uh, let's see what folks are hearing in the space. Uh, today. I was gonna talk a little bit about, uh, it's not new news, but we hadn't covered it and it was. Uh, sort of along the lines of some of the stuff we've spoken about before, uh, which is Google's universal translator, uh, tool. They've, they've not, it's not quite available yet because it's, you know, it's deep fake territory and they want it to be safe.

They want it to have watermarks, and I think that's the, that's the right approach from a big company. They can't just toss this thing out there and create massive chaos around the world. But basically what it'll do is it'll take anything that you're saying in one language, translate it into another language, but it also.

Change the way your mouth moves so that it looks like you are speaking in that other language. So, uh, that's gonna be completely bonkers. And I think on a previous episode we talked about how. These sorts of technologies are gonna unleash content onto the world that, you know, has always been there, but siloed in different languages.

And people don't wanna do, you know, uh, subtitles and they don't wanna see, uh, horrible dubbing experiences. It's going to take the amount of content out there available to you and, you know, a hundred exits like that, even though it's not new content, it's just content that you weren't really accessing, accessing before.

Uh, I thought that was. Pretty cool. It was just like one of the 50 things that they announced last week, but pretty huge news in, in my estimation. Ethan, what are you seeing out there?

Ethan: Um, yeah, a little bit extension on the testimony we talked about yesterday. Um, o one of the congressmen had mentioned a constitutional AI and I think it was very interesting they were in the weeds of that, um, and kind of understood that term.

I remember on some of the initial discussions and calls with Anthropic, um, of who of course made Claude, they kind of have a goal and initial work towards building their own like constitution within their large language models. And it's a different approach versus doing something like, Human feedback and trying to align the model for every single possible use case and how should it respond?

Um, so more philosophical and more in regards to regulation and how we continue to go down this path, but also affecting how we build these models. Um, Like I said, theoretical right now, but interesting the fact that, you know, ouris may have their own constitution at one point.

Farb: Yeah. I sent Sam a little congratulations text yesterday.

I thought he did a bang up job and wanted to make sure he, he knew that he, he, he said, thanks, uh, Connor, what are you seeing out there?

Conner: Well, you think kind of stole mine there, but Yeah, I was gonna comment on the Constitution, on the constitutional also, because it's kinda interesting that word is kind of like from Anthropic.

So the Congresswoman, or I think it was congresswoman throwing, throwing it out there, it's just kind of bit like, uh, who told you to mention that? But

Farb: it's funny, one of their staffers, uh, one of their staffers probably listens to AI daily, I imagine. You know, probably getting most of their information. I'm not sure how else they'll be,

Conner: so well inform as they were yesterday.

Farb: Yeah, absolutely. All right, some cool news today. Big news as usual. And uh, we'll see you tomorrow with more in AI Daily. See you tomorrow, guys. Have a great day, Al.

Join us on this episode of AI Daily as we discuss three exciting news stories. First, we delve into Sam Altman's testimony in front of Congress, where he discusses the future of AI regulation and its impact on the economy. Then, we explore Quora's Poe API, a groundbreaking web browser for LLMs that allows developers to bring their own language models to the platform. Finally, we cover Apple's latest accessibility announcements, including live speech and personal voice advancements. Tune in to gain insights and discover the intriguing developments in the world of AI.

Main Take-Aways:

* Sam Altman's testimony:

* Sam Altman testified in front of the Senate Judiciary subcommittee on Privacy, Technology, and the Law about AI regulation.

* He emphasized the importance of AI safety and the future of the economy with AI.

* Altman's approach of being open and accessible to lawmakers was praised.

* Senators expressed surprise that technology was actively seeking regulation.

* The hearing focused on past mistakes in technology, such as social media, and the need for good regulation.

* Quora Poe API:

* Quora introduced the Poe API, positioning itself as a web browser for language models (LLMs).

* Poe aims to allow developers to integrate any type of LLM, including custom models, into their applications.

* The API offers features like language chaining, monetization, and human feedback for reinforcement learning from humans.

* It provides a one-click replica for easy API usage and built-in integrations with LLM frameworks.

* The focus is on enabling developers to bring full LLM experiences to users, not just plugins.

* Apple's latest accessibility announcements:

* Apple announced advanced speech accessibility features, including live speech and personal voice.

* The focus is on making AI technologies accessible and beneficial for people with disabilities.

* Users can train their voices on their devices, creating custom voice models for communication.

* The Magnifier app allows users to point at objects and have the labels or buttons read aloud.

* These accessibility features leverage Apple's on-device machine learning capabilities and are expected to roll out later in the year, likely with the next OS release.

Links to Stories Mentioned:

* Sam Altman Testimony

* Quora Poe API

* Apple’s Announcements

* ChatGPT Fund

* Suhail’s “Boundless” Song

Follow us on Twitter:

* AI Daily

* Farb

* Ethan

* Conner

Transcript:

Conner: Good morning. Good morning. Welcome to another episode of AI Daily. We got three pretty great stories for you guys today, starting with first we have Sam Altman's testimony. Uh, this morning in front of Congress, the Senate Judiciary subcommittee on Privacy technology and the law interviewed him this morning, uh, parking a lot about AI regulation.

He was joined by a professor at from NYU. And also by, uh, Christina Montgomery, IBM's Chief Privacy and Trust Officer and the Senate, and Sam Altman and the other representatives were all extremely concerned about the future of AI safety, the future of the economy with ai. Farb, Any

Farb: thoughts? Uh, you know, I think Sam has been giving a masterclass in how to do this stuff correctly.

The bottom line is people want to know who the leaders are that are building and controlling these massively powered technologies, ma, massively powered technologies. Obviously the politicians want to know who these people are. The politicians want to be understood by their constituents as caring about these things, taking the steps to get these leaders in front of them to speak.

And Sam is. You know, getting ahead of the story just about every time and a, a real blueprint for other tech leaders. And clearly he's watched folks in the past, other big tech leaders from big companies who've sort of had to be, be dragged out in front of the Senate who've had to sort of been, be pulled out from, you know, hiding behind the magical curtain.

Sam is doing it out in public, uh, saying, here I am. I'm happy to speak about it. We should regulate this, and it's really working. And I, I say kudos to him and, and keep doing it.

Conner: Yeah. A couple of the senators really commented on that, how surprised they are that this is the first time that new technology has really asked to be regulated.

We've seen social media in the past section two 30, there was essentially a waiver on technology and any. Being held liable.

Farb: This is the part that I think he's reading it correctly on a lot of folks in the past. I think were afraid to do that because they're like, well, if I put myself out there, they're gonna start controlling me and start telling me what I can do, and all of a sudden, my business is not going to be as valuable.

I don't think that's the right approach. Sam's approach has been these people wanna see your face. They wanna hear your voice. They wanna know that you're a real accessible person that's willing to participate in the social contract that we have with each other around how we, you know, live and how we engage with these cool new technologies.

And, and that's the right call. Just making yourself available goes a long way.

Conner: seems to have a lot more goodwill in front of the Senate than past technology has. We've seen with social media, with meta, the entire, the entire. Duration of the hearing, they're really talking about their past mistakes in nuclear, in the genome project and social media.

They're really talking about their past mistakes and what they can learn from that, and I think Sam was really taking good stance, Ethan.

Ethan: No, I agree. I think you both nailed it. Um, at the end of the day, the, our elected officials want to do better this time. Um, and you, you pointed out very strongly how this is one of the first times that technology's asking to be regulated and Congress is open to that and they know the mistakes of the past and they want to be involved in this process.

I do think it's, you know, uh, very interesting how most of the articles coming out about this do still point at the. You know, dangers of AI and talk about how bad this can be. But if you listen to the testimony, it was actually very engaging, very thoughtful. Each congressman and each person who spoke and Sam themself talked about the positives of this technology, how it's gonna benefit people, how it can benefit.

Creatives, how the impact on jobs is not gonna be as bad as we mostly think. So if you listen to the testimony, it's actually very heartwarming to see how engaged our elected officials are, how engaged our kind of upcoming tech leaders are on this subject. And, uh, I was happy to hear it. It was very bi, it was very bipartisan.

Conner: Uh, our leaders were very engaged. Like you said, Ethan, they really seemed to know what they were going on. They talked a lot, they mentioned a couple times garbage in, garbage out, and how it related to all this of how. We need good regulation or else that's garbage in. Absolutely. Um, it really contrasts with eu, like the EU news we saw yesterday of them trying to crush open source and it's really a contrast there.

So definitely. Okay. Our next story then is Quora's Poe API. Quora announced the Poe API where they're really trying to take the angle of being a web browser for LLMs. They're really a centered piece. Po if, if you guys have used the app, it's pretty great. Yeah. Um,

Ethan: Ethan, what do you think? Yeah, Poe's been very popular for people who wanna use philanthropic and Claude and some of these other models.

And I think the most interesting thing to me here is. You know, unlike, uh, Bing or possibly Bard or even chat, g p t plugins, pos, really, I think you nailed it on a web browser almost that PO is letting you bring any type of model. So any l l m you want, if you built your own custom, l l m, if you built a whole application on top of another LLM.

They want to make that available within Poe and not just a plugin to ChatGPT, for example, but really your entire application, your entire business per se, as a custom LLM with custom features for users that they want to use. Put directly in the distribution funnel of po. So different than a plugin, bringing the whole l l m experience to someone.

So I think it's a, it's a new angle of people like using Poe for Claude, and I'm excited to see what kind of startups and developers deploy as a full l lm and not just a plugin.

Farb: Yeah. You know, as a, as someone who's not a full-time developer myself, I love the fact that they have a one-click replica, uh, that lets you fire up the.

The API and a demo and just start using it. It's the sort of thing that, you know, I have the time to actually go in there and do and start engaging with things instead of having to go to GitHub and, you know, not like it's a lot, a lot of work, but even saving 30 minutes, saving 15 minutes and just let you spend some time in the middle of your day actually interacting with the code instead of setting up, you know, where you wanna interact with it.

Getting your own replica built, uh, is really nice.

Conner: Absolutely. Yeah. It looks like they have built in integrations into Lang Chain. LA Index looks like they wanna bring monetization in the future. They have really all the features that you see in ChatGPT, but you only have to bring your own language model.

They give you, uh, human feedback that you can work with for RLHF. So between the alternatives of taking the open source for out and forcing and forking something like a HuggingChat or building for Poe, we'll see where developers go, but yep. Where's

Farb: the one click replica for every, you know, repo on GitHub.

Conner: Exactly. There's code spaces, but yeah. Yeah. Okay. Next up we have the Apple announcement. Apple announced they have live speech and personal voice, advanced speech accessibility. Uh, they're really taking the angle of it as an accessibility tool, but this is essentially an 11Labs that works on your phone.

It, you can train your voice on it, and then now you can just type in what you want it to say, and then it produces the audio sound just like you. Ethan, what are you, what are you thinking?

Ethan: Yeah, def, definitely Apple is going really on the accessibility, um, and routes. So bringing these kind of, I guess, next generation technologies and technologies, some people are still worried about, but instead of them putting them all over iOS, they're focused on, you know, accessibility and disability features right now.

The coolest one I saw was you can actually train your own voice model on your device now, so it's something you would use on 11 labs or the computer. Now you just recorded on your phone and you have your own voice model for times, you. Either don't want to talk or for some people can't talk. So Apple focusing on, you know, accessibility for these latest generation models of text and images and audio, I think is probably the right path for them right now.

Farb, this

Farb: is classic Apple, getting the technology out of the way. And talking about what it can actually do for people. So instead of saying, Hey, AI, AI, AI look at all this AI we're doing. They're just quietly doing ai. Not only that, they have ML chips on your device. You know, there are hundreds of millions of Apple devices all around the world, all with insanely powerful machine learning chips.

On the device. What's cool about this thing is like some of the other technologies, like 11Labs, you give it 15. Descript does this too. You give it about 15 minutes of your voice and it will train on that. Apple is doing it on device. It's private, it's secure. Apple doesn't even want access to your, to your voice.

It's really, really powerful. A couple of other cool things that they had in there. One, you'll be able to point the magnifier app. At a, say a, uh, washing machine, and you can point your finger on the washing machine to one of the labels or one of the buttons, and it'll read out the label of the one that you're pointing to.

Yeah. This again, happening on device, so you know, you can see Apple. Apple's great at doing this. They, they build things up over years and they slowly bring them together until they merge them into something. So you can see that the a AR glasses that Apple is expected to come out with later this year, they've been building the apps and the technologies that will power those glasses for years already.

Obviously, all these cool things that you're seeing in the Magnifier app. Like being able to point to something and having it read it for you is gonna be in the AR glasses at some point. Absolutely, absolutely. So Apple just, you know, doing their thing quietly, crushing it and, you know, maybe quietly the biggest AI company, uh, in the world, but I guess just about anything Apple does, they're the biggest company in the world doing it.

Uh, very, very cool to see. Thank you. Tim, apple.

Conner: Ethan. Is this ready now, tomorrow? When do we know, does Apple have this out already or?

Ethan: I think it's gonna start rolling out towards the end of the year, I believe is what they said.

Farb: I think typically what Apple does on these sort of accessibility announcements, and it's actually, I think, cool that they're doing it, uh, in the accessibility world first because as strange as it may sound in a world where you know, every little bit helps, it doesn't have to be perfect to be helpful.

So in a world where, you know, this is making meaningful changes in people's lives, they get it. The technology isn't perfect, it's not the smoothest perfect consumer experience available, but that's okay because it's really being used by people who need it. Uh, so typically they'll announce something like this on this, you know, world accessibility day.

Um, and then they'll sort of release it in the next os. So I would expect to see these in September. So October when, uh, the new OS rolls out.

Conner: Exactly. Okay guys. Well we had three pretty exciting announcements today. We had Sam's testimony, we had the Quora Poe API, and then of course we had Appless announcements.

So what have you guys been seeing far?

Farb: What have you seen? You know, uh, I'm a big basketball person. Played basketball my whole life. Was pretty obsessed with it when I was younger, so it was cool to see HoopGPT, uh, Maybe, maybe it's not new, but it seemed new when I, when I, when I was reading it, where it's basically allowing you to chat about all things basketball, ask for, you know, natural language versions of getting stats from people.

But one of the cool things that seemed to do is it would, for example, you could say, you know, give me a image of every spot on the floor that LeBron, you know, hit a basket against the nuggets. And so it'll generate an image where you'll see. All the shots, you know, where, where LeBron hit, hit, hit the basket.

Pretty cool stuff to see. Obviously we're gonna see this in every discipline, uh, that we haven't, and keep seeing it, uh, reiterate on in the disciplines that we have seen it. So, uh, just a little basketball news ache,

Conner: plugin, or is

Farb: that It's, it's its own thing. Yeah.

Conner: Very cool. Very cool. Ethan, what did you say?

Ethan: Yeah, I saw the ChatGPT fund on Twitter. Uh, they kind of kicked off. They put 50 grand into a fund and they're gonna let ChatGPT go on and see if it can make money. Um, it's from the team at Autopilots. Um, actually a really cool team. They're also behind the Nancy Pelosi tracker. Um, so you can. Follow her whole stock portfolio and copy trades and kind of make money that way.

And yeah, it's fun that they're making this chat. B d fund, it's been blowing up on Twitter. Um, cool to watch. Great team. Um, and yeah.

Conner: I like it. Uh, me, I saw that Boundless was a hip-hop song that Suhail made, uh, lyrics were written with ChatGPT AI to change his voice. And the bass instrument was Google's MusicLM.

So, I know we talked about before that the music, a LM only made short segments, but I assume he just made a bunch and then strung them together. Yeah. Um, I listen to it. It's definitely like a, it's like a SoundCloud quality wrap. Like I like it. Um, yeah. Pretty interesting. Very cool. See where music's gonna go.

So thank you guys. Great episode today. Um, we'll be back tomorrow. Have a good one everyone.

Farb: Thanks. Bye.

In this episode of AI Daily, we explore the groundbreaking SimpleRecon algorithm for 3D scene reconstruction. Discover how researchers have achieved faster and more accurate reconstructions using 2D convolutions, making it cheaper and deployable across various applications.

Secondly, we discuss CloudFlare Constellation, an exciting development from CloudFlare. Learn how this new platform integrates seamlessly with CloudFlare Workers, enabling easy deployment of image and language models for inference. Discover the impact this has on startups and developers, offering a simplified and cost-effective solution for running AI models.

Lastly, our third story is EU's “AI Act”, which focuses on open source developers and their responsibilities. Delve into the potential implications of these regulations, including fines and revenue percentage for models available to European consumers. Gain insights into the ongoing debate between open source and AI regulation as we explore the EU's approach.

Mentioned in Video:

SimpleRecon

CloudFlare Constellation

EU’s Regulation Announcements

ChatGPT Plugins/Browser

Follow us on Twitter:

AI Daily: https://twitter.com/aidailypod 

Farb: https://twitter.com/farbood 

Ethan: https://twitter.com/ejaldrich?s=20 

Conner: https://twitter.com/semicognitive?s=20

Join us on AI Daily as we dive into the latest breakthroughs in the world of AI! In today's episode, we discuss a game-changing update in token limits with Anthropic Claude 100K Context. We also explore the fascinating advancement of StabilityAI’s Animation SDK and its implications for creatives and filmmakers. And finally, we cover Meta AI's exciting advertising innovations, revolutionizing generative AI in the ad space! Tune in now for all the AI news you need to stay ahead!

Mentioned in This Video:

Anthropic Claude 100K Context

StabilityAI Animation SDK

Meta AI Announcements

Google’s MusicLM

Elicit

Nivi’s Twitter (Tweet on Naval & Cluade+)

Follow us on Twitter:

AI Daily: https://twitter.com/aidailypod 

Farb: https://twitter.com/farbood 

Ethan: https://twitter.com/ejaldrich?s=20 

Conner: https://twitter.com/semicognitive?s=20