Good morning. It’s Friday, January 23rd.

On this day in tech history: In 1974, Marvin Minsky published his memo on Frame Theory. Frames were the first major attempt to give AI "common sense" by using structured data to describe stereotypical situations, like entering a room. This work bridged the gap between linguistics and symbolic AI, directly influencing the rise of 1980s expert systems and the foundations of object-oriented programming.

In today’s email:

• OpenAI wants a cut of whatever you build with its AI 

• Google races toward emotionally intelligent AI 

• Apple is replacing Siri with 'Campos' and building an AirTag-sized AI wearable 

• 5 New AI Tools

• Latest AI Research Papers

You read. We listen. Let us know what you think by replying to this email.

OpenAI wants a cut of whatever you build with its AI 

OpenAI is testing a pricing model that would take a cut of whatever you build with its AI. Instead of charging purely for API calls, the company wants a percentage of "AI-aided discoveries" if you commercialize them through patents, licensing, or product sales. They're calling it a "technology empowerment fee."

The plan is to align incentives with customer success and fund deeper research. However, critics warn of potential disputes over IP and the legal chaos over who owns what, how to prove AI's contribution, and whether this kills adoption before it starts. OpenAI has not yet confirmed implementation details.

The revenue play seems to be working. CEO Sam Altman announced that OpenAI added over $1 billion in annual recurring revenue from API customers in the last month alone. The company’s commercialization blitz is accelerating everywhere. It recently launched ads in ChatGPT using cost-per-impression pricing instead of cost-per-click and dozens of advertisers are testing campaigns under $1 million each. Ads go live in February for free users and in the new $8 subscription tier. Self-serve advertising tools are under development.

The company also shipped an updated Atlas browser with tab groups and an "auto" mode that routes queries to ChatGPT or Google Search depending on which will perform better. Windows and mobile versions are in development. The advertising initiative, led by Applications CEO Fidji Simo, could provide revenue streams to support OpenAI's reported fundraising efforts seeking at least $50 billion from Middle Eastern sovereign wealth funds, with Altman already engaging state-backed investors in Abu Dhabi.

To address infrastructure demands, OpenAI launched the "Stargate Community" program, funding its own electricity through solar-plus-storage in Wisconsin, project-financed batteries in Michigan, and dedicated Texas grid buildouts, alongside closed-loop cooling.

Even governance has become political terrain. After backlash over hiring a former Democratic policy adviser, OpenAI’s CMO publicly rejected claims of ideological uniformity, citing Republican leadership and a $25 million MAGA donation from co-founder Greg Brockman. Read more. (paywalled)

Google races toward emotionally intelligent AI 

Google DeepMind acquired Hume AI's CEO Alan Cowen and top engineers to build voice interfaces that read human emotion. Hume's tech does context-aware speech synthesis and affective recognition, AI that detects emotional nuance in real time. DeepMind will fold it into Google Assistant and Gemini.

Google also launched "Personal Intelligence" in AI Mode for U.S. subscribers, letting Gemini pull from Gmail, Photos, Search, and YouTube history to personalize answers. The AI processes prompts in real time using metadata, not raw data. It's an opt-in with granular controls. The bet is an ecosystem intelligence that connects context with decisions.

DeepMind is hiring a Chief AGI Economist to report to co-founder Shane Legg. Based in London, the role will study how AGI could reshape labor, markets, and finance, underscoring that Google sees AGI as near enough to demand economic planning now. Read more.

Apple is replacing Siri with 'Campos' and building an AirTag-sized AI wearable 

Apple is rebuilding Siri and prototyping AI devices to go with it. The company is developing Campos, a system-wide chatbot that replaces Siri in iOS 27 and macOS 27 this September. It handles voice and text, integrates directly into Mail, Music, Photos, and Xcode, and lets users search files, generate content, and interact with personal data through natural language. Apple will preview it at WWDC in June.

Apple is also building an AirTag-sized AI wearable with dual cameras, three mics, a speaker, and wireless charging. It's designed to sense surroundings, though it's unclear what it does that the Apple Watch doesn't. Launch could be as early as 2027.

A third device is an AI Home Hub with a robotic swiveling display and speakers. It detects presence and rotates to face users for conversational interactions powered by the new Siri. Desktop and wall-mounted versions are in the works for spring 2026. The push shows Apple embedding AI across software, wearables, and home devices using system-level integration as the core advantage. Read more.

• Cursor 2.4 adds parallel Subagents and AI image generation with Nano Banana Pro

• Anthropic releases full rulebook controlling Claude's behavior at Davos 

• ElevenLabs made an AI album with Liza Minnelli and Art Garfunkel to show off its music generator

• Analysts, managers, and consultants? Not anymore—Deloitte modernizes titles for the AI era

• Microsoft’s VibeVoice-ASR transcribes 60-minute audio in one go

• Watch: Elon Musk, who long dismissed the WEF, tells Davos robots are coming—and they’ll surpass people by 2027

• Amazon’s One Medical launches an AI health assistant that manages your care 24/7

• AI models are getting sadder, and their music playlists quietly prove it

• Watch: Inside the $5.6B startup building robot brains

• Sen. Markey presses OpenAI and rivals on AI ads, privacy, and protecting young users

• Todoist’s new AI lets you speak your to-dos and watch them organize in real time

• Jerry Tworek quits OpenAI, warns AI innovation is trapped by metrics and money

• KREA AI launches real-time edit for real-time complex image editing

• Runway’s Gen-4.5 AI fooled 90% of viewers. Can you spot reality from deepfake?

• Alibaba open-sources Qwen3-TTS, bringing state-of-the-art voice cloning to everyone

• Over 100 fake citations slip through peer review at top AI conference

• Trump says AI race is no longer about who has the best AI Models, it is about energy

• David Sacks says overregulation and pessimism are America’s biggest obstacles in the AI race

• Spotify’s new AI lets you describe a feeling, and it builds the playlist for you

• Are AI agents ready for the workplace? A new benchmark raises doubts

• Developers can now join Vercel’s v0 waitlist ahead of its upcoming release

• Railway secures $100 million to challenge AWS with AI-native cloud infrastructure

• Google just put free SAT practice tests and a Google Classroom integration directly into Gemini

• YouTube will let creators make AI clones of themselves to pump out more Shorts

• Microsoft is turning Foxconn's failed factory site into 15 AI data centers

• Meta launched a grant program awarding up to $200,000 to groups building useful applications with its AI glasses

• Microsoft released Rho-alpha, an AI that controls dual-arm robots using language and tactile feedback

• Someone built a complete Pokémon RPG system to gamify productivity with combat formulas and gym battles

• Comic-Con banned AI art from its art show after artists called out the policy on social media

• Ollama now lets you generate images locally on Mac with two new AI models

• Meta's new AI lab delivered its first models internally while the CTO says GPT-4 to GPT-5 gains barely matter

• FlashLabs released a voice AI that outperforms humans by 11%—and you can download it right now for free

• Watch: Rifle-mounted robots made their debut at India's Republic Day rehearsal

• Volvo launches the EX60 as the world’s first Gemini-powered AI electric SUV

• Swiss suicide pod adds AI mental test to judge user fitness before activation 

• Jamie Dimon says government should have power to limit AI-driven mass layoffs

• Fractional leadership rises as AI automates execution and human judgment gains value

• Consumers can spot AI in brand videos, and it’s wrecking trust faster than marketers can keep up

• Cognition released Devinreview, a new AI tool to help you navigate through thousands of vibe-coded lines of code

📄 VJEPA: Variational Joint Embedding Predictive Architectures as Probabilistic World Models

📄 Rethinking Video Generation Model for the Embodied World ​​

📄 Being-H0.5: Scaling Human-Centric Robot Learning for Cross-Embodiment Generalization

📄 Tuning-free Visual Effect Transfer across Videos

📄 SIN-Bench: Tracing Native Evidence Chains in Long-Context Multimodal Scientific Interleaved Literature 

Thank you for reading today’s edition.

Your feedback is valuable. Respond to this email and tell us how you think we could add more value to this newsletter.

Interested in reaching smart readers like you? To become an AI Breakfast sponsor, reply to this email or DM us on 𝕏!