chat

Welcome, humans.

Meta launched a new ad campaign arguing that AI will help ordinary people “reach their full potential” and spread technology’s benefits to everyone.

That is a soothing message from the company whose last world-changing product taught your uncle to comment “AMEN” under synthetic shrimp-Jesus photos. Meta is betting on people again. Please keep the receipts this time.

Here’s what happened in AI today:

😺 ChatGPT Health can connect medical records and Apple Health.
💡 Musk said AI may surpass humanity within five years.
📰 Google found workplace AI adoption broad but shallow.
📰 Patreon cut 20% of its workforce amid AI upheaval.
🎓 Audit an AI agent’s decisions instead of its code.

…and a whole lot more that you can read about here.

P.S: We just launched a robotics newsletter! Sign up for it here.

😺 ChatGPT Health Can Now Read Your Medical Records

The hardest part of asking a health question is often reconstructing your own body first. One answer lives in a patient portal, another in Apple Health, and the lab result you need is buried under three password resets.

OpenAI is turning ChatGPT into the layer that connects those fragments.

Here’s what happened:

Health in ChatGPT is rolling out to logged-in U.S. users age 18 and older across Free, Go, Plus, and Pro plans on web and iOS.
You can connect Apple Health and supported medical records, then ask questions grounded in medications, visits, labs, sleep, activity, and other information.
OpenAI says connected health data and conversations that use it will not train its foundation models or target ads.
ChatGPT is meant to explain and organize health information, not replace professional care.

OpenAI says more than 300M people ask ChatGPT health questions each week. The new system can compare lab results over time, summarize changes since an appointment, or connect exercise and sleep patterns without making you upload the same files repeatedly.

How to try it:

Open Health from ChatGPT’s sidebar or More menu.
Select “Get started,” then connect Apple Health or a supported medical-record provider.
Review medications and conditions, since synced records can be incomplete or outdated.
Ask a narrow question, such as “What changed in my lab results since my last appointment?”

Why this matters: ChatGPT is moving from answering generic health questions to interpreting your personal history. That could make it far more useful for appointment prep, translating medical language, spotting changes across records, or remembering the question you meant to ask your doctor.

It also raises the cost of a confident mistake. A polished answer grounded in real records can feel more trustworthy than a generic chatbot response, even when the underlying judgment is wrong. The safest use is preparation: organize the evidence, explain unfamiliar terms, and generate better questions for a clinician.

Our take: OpenAI built meaningful privacy controls around Health, including extra encryption, per-use permission prompts, and a promise that connected data will not train models or target ads. The harder problem is trust. Health information is messy, incomplete, and consequential, so the product will be judged less by how polished it sounds than by how often it knows when to slow down.

The next healthcare interface may be the assistant sitting between you and every patient portal.

FROM OUR PARTNERS

FREE Claude AI Mastery Workshop

Claude is currently the most powerful tool of 2026. It’s been launching new features every week: Skills, Connectors, Cowork, vibe coding. Yet almost no one knows how to use them.

Our expert mentors condensed 800+ hours of Claude research and real-world practice into a focused 16-hour curriculum. Join the 2-Day Claude AI Mastery Workshop: a live, end-to-end deep dive into Claude plus 10+ AI tools, LLMs and workflows. SIGN UP.

Learn to master Claude’s Chat, Cowork and Code modes; set up Skills, Connectors and Plug-ins to automate your desktop, Notion and files; and vibe code apps and dashboards without writing code.

🧠 Saturday & Sunday, 10 AM–7 PM EST. Register NOW!

🎓 AI Skill of the Day: Run a Decision Audit Before You Merge

AI can write thousands of lines in minutes. Reviewing every line defeats the point, but blindly merging its work is how a codebase becomes digital lasagna. You ain’t Garfield.

Victor Taelin’s rule is simpler: audit the AI’s choices, not every line of code. Coding agents usually execute a concrete plan well. The danger appears when the task is underspecified and the agent quietly chooses an architecture, shortcut, or assumption for you.

Use this workflow:

Lock the important decisions before execution: desired behavior, constraints, architecture, and what “fixed” means (if fixing something).
After the agent finishes, ask it to list every meaningful choice it made, especially anything it was unsure about.
Review that short list instead of the full code diff. Correct weak assumptions, or bad choices, then have the agent revise the implementation.
Run a “pride gate” (per the top commenter on Taelin’s post): ask whether the coding agent is proud of the branch and would stand behind it in production.

An AI with no ego is finally useful! It will often admit which shortcuts, edge cases, or temporary fixes still bother it.

This turns code review into a conversation about judgment and higher order decisions, rather than a thousand-line scavenger hunt to figure out wtf Claude just wrote…

Before I merge this work, audit the decisions you made:

1. List every meaningful decision, assumption, shortcut, or interpretation you made while completing the task.
2. For each one, explain why you chose it, what alternatives you considered, and how confident you are.
3. Flag any choice that fixes the immediate example but may not solve the underlying problem generally.
4. Identify any edge cases, technical debt, or temporary fixes that remain.
5. Pride gate: Are you proud of this branch? Would you confidently stand behind these changes in production?

Be candid. If the answer to either question is no, explain exactly what should be improved.

Do not change the code yet. Wait for me to review your decisions.

Have a specific skill you want to learn? Request it here.

🍪 Treats to Try

Claude voice mode helps you reason through harder problems with Opus or Sonnet, use connected tools like Gmail and Slack, and switch among more languages (free with Haiku; expanded models from $20/mo).
Google Selfie Video gives you a backup way into your Google Account by comparing a short live video with one you securely recorded earlier (free).
FLUX 3 is Black Forest Labs’ first model to move beyond its popular open-weight image generators, creating and editing images, video, and audio in one system (early access; pricing not public).
Runway Media Router automatically chooses the best image, video, or audio model for your request based on your preferred quality, speed, and cost (from $0.01 per credit).
Screenpipe records your screen and audio locally, making past work searchable and giving AI agents long-term context about what you have done (free core; hosted app from $25/mo).
Cursor Router sends each coding request to the best-fit model, with Cost, Balance, and Intelligence modes and up to 60% savings in Cursor’s tests (Teams from $40/user/mo).

New from The Neuron: AI Explained podcast:

Matthias Unberath of Inner Logic on The Neuron AI Explained

📰 Around the Horn

Google’s own data showed workplace AI use had reached 68% of U.S. occupations, but covered only about 21% of tasks in a typical job.
Patreon laid off 20% of its workforce as CEO Jack Conte said AI had fundamentally changed how tech companies work and organize.
DeepSeek founder Liang Wenfeng resurfaced with a translated transcript expanding on the long-term research ambitions behind his previously reported $10B AGI bet.
Anthropic expanded Claude voice mode to Opus and Sonnet, while OpenAI released voice models designed for more natural live conversations.
SemiAnalysis argued OpenAI and Anthropic had separated into a “two-horse race” at the frontier.

Want absolutely EVERYTHING that happened in AI today? Read this.

FROM OUR PARTNERS

Watch how owning AI deployment expands your career

Missed the live roundtable? Watch three teams share how they made AI ownership their job, now on-demand.

Watch On-Demand

💡 Intelligent Insights

The Economist interviewed Elon Musk, who predicted AI could exceed humanity's combined intelligence within five years and said humans may no longer be in control within ten… and yet he’s optimistic about this, because he thinks we can’t stop it progressing, so he says we should enjoy the ride…
Nick Saraev (featured above) was deeply disturbed by Claude’s new voice mode, arguing that the real danger is not AI becoming more human, but humans gradually sanding away their individuality to become easier for machines to understand.
Benjamin Marie estimated that Alibaba’s planned 2.4T-parameter Qwen3.8 would be nearly impossible to run on a normal PC, even after aggressive quantization.
Poolside cofounder Eiso Kant explained how its Model Factory speeds experimentation, why Laguna S uses open weights, and how iteration speed could shape the race toward AGI.
The Context Window Lie unpacked why million-token context windows provide temporary working space rather than durable memory; this is unsolved AI issue #1, followed by safety guardrails, hallucinations, and the so-called “continual learning.” IMO, solve one, and you solve them all.
Anthropic’s Thariq Shihipar demonstrated an agentic engineering workflow built around iterative planning and coding agents that revise their approach while working.
Neil deGrasse Tyson and Jaron Lanier examined Lanier’s argument that AI is less an independent intelligence than a vast human collaboration (I couldn’t agree with this more), reframing black-box safety, data rights, and the risk of automated slop.
Stanford economist Erik Brynjolfsson said AI’s economic payoff should become unmistakable within three to five years, but the next decade’s outcome depends on reskilling, shared prosperity, and humans retaining the question-setting and judgment work.