In partnership with

Good morning. It’s Friday, January 9th.

On this day in tech history: In 2010, the Neural Turing Machine concept began circulating internally at DeepMind, exploring neural networks with differentiable memory. Long before transformers dominated, this work seeded ideas around external memory, attention, and algorithmic reasoning now resurfacing in tool-augmented LLMs.

In today’s email:

• OpenAI is running a two-track healthcare play for consumers and hospitals

• Google overtakes Apple, pumps Gemini AI into Gmail and Classroom

• Claude Code 2.1.0 brings hot-reload, multilingual output, and real-time workflow control

• 5 New AI Tools

• Latest AI Research Papers

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

OpenAI is running a two-track healthcare play for consumers and hospitals

OpenAI is stepping up in healthcare with a two-pronged strategy.

ChatGPT Health, a new tab inside ChatGPT, lets users link medical records and wellness apps like b.well, Apple Health, MyFitnessPal, Peloton, and Weight Watchers. It provides context-aware insights on lab results, activity, and nutrition, helping users prep for visits or navigate insurance.

Data runs in a sandbox: isolated from regular chats, deletable, and excluded from model training. Context flows into Health, but health data never flows back. A limited beta is live via waitlist, with wider web and iOS availability planned.

For hospitals, OpenAI launched OpenAI for Healthcare, a HIPAA-compliant, enterprise-grade platform now live at Memorial Sloan Kettering, Cedars-Sinai, UCSF, and Stanford Medicine. GPT-5-class models handle clinical workflows, evidence-based reasoning with citations, structured prompts, and reusable documentation. Enterprise controls include SAML, SCIM, role-based access, audit logs, and customer-managed encryption.

OpenAI also acqui-hired the founding team of executive coaching AI startup Convogo, shutting down the product but bringing in applied AI talent to expand cloud and enterprise capabilities. Read more.

Google overtakes Apple, pumps Gemini AI into Gmail and Classroom

Google nailed an AI masterstroke, and it's paying off big.

Alphabet just overtook Apple, hitting a $3.89 trillion market cap to claim the global No. 2 spot behind Nvidia. Investors are all-in on Gemini's strength and Google's custom TPUs, which power massive AI workloads across search, cloud, and enterprise. The proof is in the numbers: Gemini now commands over 20% of global chat traffic, steadily chipping away at ChatGPT's share.

The Gemini AI suite is now central to multiple Alphabet products. Gmail, with over three billion users, is entering the “Gemini era” through the Gemini 3 model. AI Overviews now auto-condense messy threads into tight summaries and handle natural-language inbox queries.

Suggested Replies deliver spot-on, tone-matched one-tap responses that feel authentically you. Premium subscribers get advanced Proofread for expert grammar, clarity, and style fixes. Coming soon: AI Inbox intelligently surfaces urgent to-dos, bills, and VIP threads by inferring relationships and patterns all processed securely, no data used for training.

Gemini’s impact extends to education. Google Classroom now converts lessons into podcast-style episodes, allowing teachers to control grade levels, topics, learning objectives, speaker roles, and conversation style. Students can replay lessons at their own pace. Read more.

Claude Code 2.1.0 brings hot-reload, multilingual output, and real-time workflow control

Anthropic has released Claude Code 2.1.0, a major upgrade to its “vibe coding” platform for building autonomous AI agents and structured workflows.

Powered by Claude Opus 4.5, the update adds agent lifecycle hooks, forked skill contexts, hot-reload, session teleportation, wildcard tool permissions, and multilingual output effectively reducing setup friction and enabling real-time iteration across multi-step, modular workflows.

Developer experience sees a boost with improved terminal UX, Vim-style motions, real-time reasoning visibility, skill progress indicators, and slash-command invocation. Security and reliability are strengthened with fixes for session persistence, API context overflow, and sensitive data exposure. Claude Code 2.1.0 is aimed at power users and enterprise teams treating AI as programmable infrastructure rather than a chatbot.

The launch coincides with a $10 billion funding round valuing Anthropic at $350 billion, almost double its worth four months ago. The round, led by Singapore’s GIC and Coatue Management, complements $15 billion in planned Microsoft–Nvidia investments, including $30 billion in Azure compute. Read more.

• Nvidia CEO says AI on older GPUs is possible

• Tailwind CSS is booming, but AI slashed revenue by 80%, now Google AI steps in to help

• Grok Code surges to #1 on AI leaderboards, Musk promises one-shot coding for complex tasks

• 7 CES 2026 devices that turn everyday life into a Black Mirror episode

• Only Claude refused to play along when tested with a fake idiom by an AI researcher

• AI coding assistants are quietly getting worse—GPT-5 and newer models now produce “silent failures”

• Microsoft’s Copilot Checkout goes live with PayPal and Stripe powering in-chat purchases

• Linus Torvalds shuts down AI slop panic: “Documentation won’t fix bad code”

• US federal judge greenlights Elon Musk’s jury trial against OpenAI over its for-profit pivot

• LLM pioneer Zhipu AI goes public in landmark Hong Kong IPO

• Qwen3-VL-Embedding and Reranker redefine cross-modal search with open-source power for developers

• Arm Holdings launches “Physical AI” to merge robotics and automotive under one roof

• CES 2026 unveils AI boats that drive, dock, and think alongside you

• Roku CEO predicts the first 100% AI-generated hit movie within three years

• DeepSeek's AI gains traction in developing nations, Microsoft report says

• Exoskeletons are shedding their sci-fi bulk to become the next wearable tech

• Nvidia CEO calls robots “AI immigrants” and says they’ll solve the global labor crunch

• Northeastern’s new AI can predict your next move, hours before you make it

• Meta’s $2 billion Manus deal hits a regulatory snag as Beijing questions AI export rules

• Watch: This robot vacuum climbs stairs

• Microsoft’s AI push may cost 22,000 jobs even as revenue stays strong

• Sony patents AI “Ghost Player” that could finish PlayStation games for players stuck on tough levels

• Microsoft-led research shows AI can create new viruses and bypass safety filters

• Dell’s CES 2026 keynote whispered “consumer-first” and skipped the AI hype

• A new AI leaderboard reveals who wins when language models are forced to play politics

• Utah is the first state to allow AI to renew medical prescriptions, no doctors involved

• The mind behind Libratus is trying to crack Poker’s last street with AI

• Open-source video AI moves beyond APIs as LTX-2 ships full models and tooling

• Tensor’s AI-first robocar turns a vehicle into a rolling supercomputer

• Watch: Kawasaki’s new AI robot hitches a ride on a drone to fix turbine blades in real time

• NousCoder-14B matches trillion-dollar AI rivals in four days, and it’s completely open-source

• Oxford Economics suggests the AI apocalypse in jobs is mostly a corporate PR story

• Ford is bringing AI to your car with a smart assistant in 2027 

• McKinsey and General Catalyst execs say the era of 'learn once, work forever' is over

• MiroMind’s 30B AI just matched trillion-parameter rivals while costing 1/20th as much

• 𝕏’s deepfake machine is infuriating policymakers around the globe

• VC predicts 2026 will be the year consumer AI goes mainstream — and these are the apps OpenAI won’t touch

• Cursor release a dynamic context discovery for all models which reduced total tokens by 46.9%

• AI is quick but risky for updating old software, researchers warn

• This AI can predict your future illnesses from a single night of sleep 

• Watch: Lenovo debuts Qira, an AI that follows you across all your devices without ever opening an app

• The problem with AI isn’t intelligence, it’s taste. Replit shows how to fix it

📄 InfiniDepth: Arbitrary-Resolution and Fine-Grained Depth Estimation with Neural Implicit Fields

📄 DreamStyle: A Unified Framework for Video Stylization ​​

📄 SOP: A Scalable Online Post-Training System for Vision-Language-Action Models  

📄 Benchmark^2: Systematic Evaluation of LLM Benchmarks 

📄 Entropy-Adaptive Fine-Tuning: Resolving Confident Conflicts to Mitigate Forgetting 

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!