Robots tripping over soccer balls. Phones that speak your language. And chatbots that say they can feel “distress.” AI keeps surprising us but are these steps forward or just strange detours?

This week, Jeff and Annie unpack the latest mix of useful and slightly unsettling AI news. They explore how AI is creeping into every corner of life. And then comes the big question: should we treat AI like a tool or like it has feelings of its own?

We also discuss:

• Google’s Pixel 10 goes all-in on on-device AI
• Microsoft’s Excel gets a natural-language Copilot
• Grammarly’s AI grader predicts your score before your teacher does
• Perplexity’s new browser acts like a personal web agent
• Anthropic vs. Microsoft: are we blurring the line between code and consciousness?

Chapters:

00:00 – Robot Olympics: A Humorous Take on AI Progress
02:58 – AI Features in the New Google Pixel 10
05:57 – Microsoft Excel’s AI Integration: A Game Changer?
09:02 – AI in Education: Tools for Students and Teachers
11:47 – AI’s Role in Scientific Research: A New Conference
14:54 – The Ethics of AI: Consciousness and Emotional Distress
18:02 – AI in Browsers: A New Era of Internet Navigation
20:59 – Anthropic’s Claude AI: Ending Abusive Conversations
23:59 – The Future of AI: Balancing Human Needs and AI Rights

Relevant Links:

• Robot Olympics features face‑planting robot soccer and sprint fails
• Google’s Pixel 10 launches Tensor G5 chip boosting on‑device Gemini AI
• Microsoft adds Copilot function to Excel powered by GPT‑4.1‑mini
• Grammarly launches AI agents including AI Grader, Plagiarism Checker
• Claude can now end abusive chats, citing AI “welfare”

From leaked AI rulebooks to state bans on AI therapy, Annie and guest Charlie Costello dig into the messy, human side of setting limits for machines.

Annie sits down with her Upstart colleague Charlie Costello to talk about the real-world decisions shaping AI’s future. They break down Meta’s leaked chatbot guidelines—yes, the one with “romantic chats with minors” in the fine print—debate whether “deep ignorance” can make AI safer by erasing dangerous knowledge, and unpack Illinois’ surprising move to ban therapists from using AI (even while patients still can).

We also discuss:

• What Meta’s 200-page chatbot rulebook says—and what it leaves out
• Can “deep ignorance” keep AI safe without making it useless?
• Why Illinois banned AI in therapy, and what it could mean for mental health care

It’s finally here. After months of hype and “is it coming this week?” guesses, OpenAI has dropped GPT-5—and it’s not just another model update. This one’s faster, smarter, and feels a whole lot more human.

And OpenAI didn’t stop there. They also surprised everyone with their first open-weight model in over six years. Meanwhile, Google and Anthropic weren’t about to sit quietly—both rolled out big updates of their own.

In this episode, Jeff Keltner dives into the week’s AI news, including:

• The five standout features that make GPT-5 a real game-changer
• Google’s Gemini 2.5 Deep Think and its team-of-AI-agents approach
• Anthropic’s Claude Opus 4.1 upgrade
• Bold moves in AI media—from Elon’s spicy Grok videos to 11 Labs’ text-to-music magic

Relevant links:

• Grok Imagine now generates 15s videos in-app
• ElevenLabs unveils Eleven Music for full song generation
• Google’s Genie 3 builds persistent 3D world simulations
• Gemini Storybook creates illustrated, narrated kids’ books
• Microsoft unveils CLIO for real-time reasoning adaptation
• DeepMind and Kaggle launch Game Arena for model benchmarking
• OpenAI redesigns ChatGPT for emotional safety and better outcomes

What if China is teaching AI better than the U.S.? And what if your next doctor or teacher… was an AI?

This week, Jeff Keltner is joined by Dr. Ivy Lee — a physician and AI policy leader — for a sharp, grounded take on the latest AI developments.

They unpack China’s aggressive push to make AI a core academic skill, explore Meta’s vision of smart glasses as “personal superintelligence,” and raise big questions about who controls the future of learning.

We also cover:
- Amazon quietly backs a “Netflix of AI” app where users generate animated TV shows
- Microsoft Edge evolves into a real AI agent, acting across tabs and tasks
- Google’s NotebookLM now turns your documents into narrated video explainers

Relevant Links:

• China’s AI blueprint focuses on global collaboration and open access
• Alibaba drops Wan 2.2, a next-gen video generation model
• Zhipu AI’s GLM-4.5 released as fully open model with agentic abilities
• Meta bets on smart glasses and personal superintelligence
• DeepMind unveils AlphaEarth, an AI for planetary-scale monitoring
• Amazon-backed Fable launches Showrunner, the “Netflix of AI”
• Adobe adds “Harmonize” to Photoshop, blurring reality and fiction
• Microsoft Edge rolls out Copilot Mode for AI-powered browsing
• Google’s NotebookLM adds video overviews to documents
• OpenAI releases Study Mode to guide students with Socratic prompts
• Stanford’s virtual AI lab designs COVID antibodies in days
• China mandates AI literacy across universities and K–12 schools

What happens when Trump drops an AI blueprint, Google wrecks your traffic, and AI gets psychologically manipulated?

This week, Jeff and Annie tackle one of the most thought-provoking episodes yet—from policy and power to psychology and physics. They break down Trump’s aggressive AI action plan, an OpenAI vs. DeepMind math showdown, and a Kenyan health experiment where AI saves lives behind the scenes.

In this episode:

• Brain-inspired small model beats giants in reasoning tests
• Google Photos’ new remix tools and animation features
• AI-designed physics experiments that shock actual physicists
• Wharton study shows how AI can be socially manipulated

Relevant Links:

• Trump drops aggressive AI Action Plan
• OpenAI and DeepMind hit gold at Math Olympiad
• Sapient’s HRM model mimics the human brain
• AI-designed physics upgrades LIGO sensitivity
• Amazon acquires Bee, an AI-powered memory wearable
• Pew study: Google’s AI Overviews hurt web traffic
• Google Photos adds remix-style AI video tools
• OpenAI’s AI Copilot cuts medical errors in Kenya
• Wharton researchers show AI can be socially manipulated
• Meta refuses to sign EU’s voluntary AI Code

We were told AI would make us faster, smarter, more efficient. But what if it’s just making us feel productive—while secretly slowing us down?

This week, Jeff and Annie dig into the most surprising stories in AI. Popeyes used AI to launch a diss track against McDonald’s. A $3 billion deal between OpenAI and Windsurf collapsed, and Google swooped in with a $2.4 billion move of its own. A new study shows experienced developers were actually slower when using AI—yet completely convinced they were faster.

We also cover:

• Google’s MedGemma brings AI to medical imaging (even on your phone)
• ChatGPT Agents can now build presentations, shop, and run apps on their own
• BCG report reveals most companies aren’t seeing cost savings from AI
• Netflix finishes a major VFX scene 10x faster with GenAI
  
  

Relevant Links:

• Google licenses Windsurf for $2.4B after OpenAI deal fails
• Devs using AI are 19% slower, METR study shows
• Google’s MedGemma hits 87.7% on MedQA benchmark
• Self-driving chemistry lab accelerates discovery
• Runway’s Act 2 brings motion capture to AI video
• Adobe Firefly adds voice-to-sound FX for video editing
• Google Veo 3 generates video with dialogue—but buggy subtitles remain
• OpenAI drops ChatGPT Agent with autonomous actions
• BCG: Only 1 in 4 companies see ROI from AI
• Netflix uses GenAI to cut VFX time 10x
• Cloudflare blocks AI bots with “Pay per Crawl” model

What happens when AI outsmarts your doctor, charms your date, and impersonates a U.S. senator?

This week, Jeff and Annie dive into some AI stories. A fertility breakthrough gives new hope to families after 18 years of trying. Microsoft claims its AI can now beat real doctors at complex diagnoses. And in a truly unsettling twist, someone used a fake voice clone of Marco Rubio to try and access classified information.It’s not just breakthroughs—it’s questions of trust, privacy, and what happens when AI starts thinking (and strategizing) like us.

We also discuss:

• XAI’s Grok 4 Heavy: $300/month for a truth-seeking muscle machine
• Perplexity’s new AI browser that can book your meetings and browse for you
• New research shows LLMs have strategy fingerprints—and personalities

Relevant Links:
Columbia's STAR system finds viable sperm using AI
DeepMind’s Isomorphic Labs starts cancer drug trials
Grok 4 and Grok 4 Heavy benchmark results
Perplexity launches AI-powered Comet browser
Nikkei reports scientists used hidden AI prompts
Anthropic proposes AI transparency framework
AI deepfake impersonates Marco Rubio
Microsoft MAI-DxO claims diagnostic breakthrough
AI models show unique strategy patterns in games

Jeff and Annie are back with another round of “Wait, what?!” moments from the AI world. Mira Murati raises $2 billion for a stealth company no one can quite explain. DeepMind’s new robots don’t need the cloud—or a human form—to do real work. And Claude is now building apps straight from your chat window.

We also cover:

• The surprising potential of disembodied robot arms
• What AI-native customer service looks like in the wild
• The ethics breakdown in Anthropic’s latest red-teaming tests
• Big legal shifts in AI training data and fair use

Relevant Links:
Mira Murati’s $2B AI startup raises record-breaking seed round
DeepMind launches Gemini-powered robots with no cloud access
DeepMind unveils AlphaGenome for gene regulation prediction
Claude now lets you build apps inside the chatbot
ChatGPT adds integration with Google Drive and OneDrive
Verizon deploys Gemini-powered customer service AI
Anthropic’s Claude resorted to blackmail in testing
Judge rules AI training on purchased books is fair use

What if your AI twin outsold you, and ChatGPT made your brain… slower?

In this episode, Jeff and Annie explore the strange ways AI is reshaping how we think, work, and even sell. Meta’s smart glasses are getting a reboot, OpenAI’s “Projects” feature keeps leveling up, and China just pulled off a $7M livestream using two AI hosts. Meanwhile, an MIT study raises red flags about what writing with ChatGPT might be doing to our brains.

We also discuss:

• Meta’s smart glasses push with Oakley and what consumers actually use them for
• OpenAI’s steady rollout of new ChatGPT features (and their new podcast)
• MIT's SEAL framework for self-improving models — and why it matters
• A $7M AI live-stream sales event in China and the rise of AI influencers
• McKinsey’s take on why genAI isn’t delivering ROI — yet
• A new UK study on kids using AI — and how it’s splitting along private vs. public education lines

Relevant Links:
Meta announces Oakley smart glasses with AI features
OpenAI upgrades ChatGPT Projects with voice and mobile tools
OpenAI launches official podcast hosted by Andrew Mayne
MIT unveils SEAL: AI that improves itself with no humans
China’s AI models show human-like internal concept mapping
McKinsey report: Why genAI isn’t improving the bottom line
$7M Baidu livestream driven by AI avatars
China’s $1T digital human industry explained
Alan Turing Institute study on AI use in UK schools
MIT brain scan study on ChatGPT writing impact

This week, Jeff and Annie delve into the weird, wonderful, and slightly concerning ways AI is making its presence felt in our lives. Meta is throwing $15B at a new AGI dream team. Google’s Gemini wants to become your executive assistant. OpenAI is in court defending your right to delete. AI is showing up in hospitals, planning departments, even behind the Starbucks counter, and not always in the ways you’d expect.

In this episode:

• A new study reveals the real secret to better AI prompts—and it’s not what you think
• The privacy battle that’s putting OpenAI in the awkward role of data defender
• The growing tension between states and the feds over who gets to regulate AI

Relevant Links:
Google’s Gemini adds scheduled actions for productivity
Wharton study debunks prompt engineering myths
OpenAI fights NYT over user chat data retention
Meta invests $15B in Scale AI for AGI ambitions
Amazon launches AI video generator for sellers
Ohio State mandates AI fluency for all students
DeepMind’s Weather Lab beats top cyclone models
Heartfelt Technologies builds AI scanner for heart failure detection
UK and Google launch AI tool ‘Extract’ for planning reform
Starbucks tests Green Dot Assist AI at 35 stores
Trump-backed “Big Beautiful Bill” includes 10-year state AI regulation ban