😺 OpenAI found 18 rare diseases

Welcome, humans.

The strangest AI headline today was not a chatbot writing another email. It was OpenAI backing Rust maintainers because the future of AI apparently depends on the programming language your most intense engineer has been telling you about since 2018.

That is the funny part of this cycle: the flashiest AI demos still run on deeply unflashy infrastructure, from compilers to libraries to the people who keep build tools from catching fire. Somewhere, a maintainer just whispered, 'finally.' Anyway, let's talk about the other place AI is getting very real: the doctor's office.

Here’s what happened in AI today:

• 😺 OpenAI used a reasoning model to help doctors surface 18 rare-disease diagnoses from previously unsolved cases.

• 📰 Anthropic added enterprise-managed auth for MCP connectors across Claude chat, Claude Code, and Cowork.

• 🍪 fal released LTX 2.3 LoRA trainers for custom media generation.

• 💡 TestingCatalog reported that OpenAI is preparing GPT-5.6 and GPT-5.6-Pro.

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

Hey: Want to reach 700,000+ AI-hungry readers? Advertise with us! 

P.S: Love robots? We’re starting a new robotics newsletter! Sign up early here.

😺 OpenAI's latest health push made ChatGPT feel a lot closer to the exam room

The most important AI product in health might be the one people already open at midnight, when the rash looks weird and the doctor's portal is closed.

OpenAI said GPT-5.5 Instant now gives stronger health answers to free ChatGPT users, while a separate NEJM AI study used an OpenAI reasoning model to help physicians revisit rare-disease cases that had already stumped specialists.

Here's what happened:

• OpenAI says more than 230M people ask ChatGPT health and wellness questions every week.

• GPT-5.5 Instant, available to free users subject to limits, improved on health evaluations around urgent-care recognition, uncertainty, and context gathering.

• Researchers at Boston Children's Hospital, Harvard, and OpenAI reanalyzed 376 de-identified unsolved cases with o3 Deep Research.

• Doctors confirmed 18 new diagnoses after expert review, testing, and clinical validation.

That last part matters because rare diseases are brutal for families. Even after genomic sequencing, OpenAI says roughly half of rare-disease patients remain undiagnosed after extensive specialist review. The clues can be buried across clinical notes, variants, old test results, and medical literature that keeps changing.

Why this matters:

AI health tools are moving from general advice into workflow support. The rare-disease study did not let a model diagnose patients on its own. It used the model to generate evidence-linked leads, then human clinicians decided what deserved follow-up.

That is the useful near-term shape: AI as the tireless second reader, not the doctor of record. The risk is that consumers will treat a polished answer like a final answer, especially when care is expensive or slow.

The timing is also awkward in the very human way health tech always is. ChatGPT can help people ask better questions, but medical systems still decide who gets testing, who gets appointments, and who pays. Better answers are only part of the care gap for patients and families.

Our take:

OpenAI is turning health into one of ChatGPT's biggest mainstream jobs. The test now is whether it can make people better prepared for care without convincing them they can skip care entirely, especially when the answer sounds calm, confident, and complete on a stressful night alone.

FROM OUR PARTNERS

See Exactly Where Your AI Investments Should Go.

Most enterprises make AI decisions based on how they think work happens — not how it actually does. Scribe Optimize changes that by automatically detecting real workflows across your org, so leadership finally has the data to act.

Map real workflows without surveys or consultants

Surface inefficiencies leadership can't currently see

Build your AI strategy on ground truth, not gut feel

See how work actually happens

🎓 AI Skill of the Day:

Use ChatGPT as a pre-appointment prep assistant, not a replacement doctor. The goal is to turn messy symptoms, lab notes, and questions into a cleaner agenda for the human clinician.

Before an appointment, paste only the details you are comfortable sharing, then ask the model to organize them into: timeline, red flags to mention, questions to ask, and what information you should bring. The useful move is asking for uncertainty, not certainty.

I am preparing for a medical appointment. Do not diagnose me. Help me organize the information below into: 1) a short symptom timeline, 2) important context to mention, 3) possible red flags I should ask about, 4) questions for my clinician, and 5) documents or test results I should bring. If anything sounds urgent, tell me to seek professional care instead of waiting.  

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

🍪 Treats to Try

*Asterisk = from our partners (only the first one!). Advertise to 700K+ readers here!

1. *10 weeks. $30M USD in AI impact. This can be you. See how we did it. →

   Let's build it together

2. Google Home Speaker brings Gemini into a $99 smart speaker for more natural home commands and follow-up questions.

3. Apartments.com Ai lets renters search, compare, and evaluate apartments through a conversational assistant, no pricing details.

4. GitHits lets coding agents inspect the open-source packages your app depends on before they guess at implementation details (read more).

5. Daemons by Charlie Labs monitors PRs, issues, CI, docs, Slack, Linear, GitHub, and Sentry so engineering teams can keep agent-created work moving, no pricing details.

6. fal LTX 2.3 LoRA Trainers helps you train custom media-generation and editing models across video, image, and audio workflows - no pricing details.

7. Google Workspace Vids turns slides into avatar-led videos with generated voiceovers in 24 languages - no pricing details.

8. Ludeo: playable gameplay moments makes gameplay clips instantly shareable and playable through its SDK for game developers.

Learn AI From Scratch Via Our Live Stream Guide

We’ve been doing a lot of AI for Total Beginners streams this week. Here’s an ordered guide of what order to watch them in to get the most out of them:

1. Start with The 5-Step Framework to Learn AI in 2026 for the map: projects, prompting, skills, automations, and agents (article version).

2. Watch The AI Starter Kit: What to Try...and What to Ignore to pick the right first tools and skip the which subscription analysis paralysis.

3. Watch AI Skills vs Agents vs GPTs: Which One Do I Use? to understand the containers: projects, GPTs / Gems, skills, and agents (article version).

4. Watch Learn Agents in 2026 With This Total Beginner’s Guide to AI Agents & Automation for context, tools, triggers, approvals, and human-in-the-loop guardrails (article version).

5. Watch OpenAI Workspace Agents 101: Build, Run, and Scale AI Workflows for the build stage: connections, scheduled runs, templates, and useful agents without a terminal directly inside GPT.

6. Finish with Building Real-Time AI Voice Agents with LiveKit’s Ben Cherry, the advanced track on phone agents, screen sharing, speech-to-text, text-to-speech, real-time models, debugging, observability, and product infrastructure (article version).

By the end, you’ll know when to use each one.

📰 Around the Horn

• Boston Dynamics Spot robots were deployed for selected FIFA World Cup security and inspection work.

• Google rolled out Gemini-powered Gmail summaries globally across free, paid, and Workspace accounts on Android, iOS, and web.

• Apple warned that AI data-center demand for memory chips could make iPhones, Macs, and other devices more expensive.

• Anthropic added centralized enterprise auth for MCP connectors, starting with Okta beta support.

• Cerebras previewed Google's Gemma 4 multimodal model on Cerebras Inference at more than 1,500 output tokens per second.

• Amazon was reportedly in talks to sell Trainium3 AI chips for other companies' data centers.

• FERC ordered regional grid operators to justify or overhaul data-center power-connection rules within 60 days.

• Anthropic is “very confident” Claude Fable 5 and Mythos 5 will become available again “in the coming days,” according to Managing Director of International Chris Ciauri, though the models remain officially unavailable for now after a U.S. government directive forced Anthropic to block access.

• Google⁠ and Microsoft⁠ announced Agentic Resource Discovery, an Apache 2.0 open specification for publishing, discovering, and verifying AI tools, skills, agents, and other resources across federated registries, with GitHub’s Agent Finder launching as an implementation for Copilot. 

• Perplexity⁠ launched Brain for Computer, a continuously learning memory/context-graph system in Research Preview for Max subscribers that starts future tasks with project context and is reported to improve answer correctness by 25%, recall by 16%, and history-dependent task cost by 13%

FROM OUR PARTNERS

The AI tinkerer is now the AI operator

Dell Pro Max with GB10 gives your builder a serious desktop for experiments, demos, and workflows, powered by NVIDIA Grace Blackwell architecture, NVIDIA DGX OS 7, and 128GB memory.

See Dell Pro Max with GB10.

💡 Intelligent Insights:

• OpenAI: rare childhood disease diagnosis showed how a reasoning model helped clinicians surface 18 confirmed diagnoses from 376 previously unsolved pediatric genetic-disease cases.

• 12 Grams of Carbon: HTML is all you need pushed coding agents to make graphics and visual media with HTML as the core creative medium.

• Elena Verna: the Mom-and-Pop SaaS era argued that AI makes tiny software markets viable by letting domain experts turn lived expertise into profitable products.

• 404 Media: If AI is sentient, so is Age of Empires II covered Adrian de Wynter's argument that we anthropomorphize interfaces too quickly, even when the same primitives can be built inside a strategy game.

• Reuters⁠ reported that a Stanford-cited “intelligence per watt” study found local small language models running on PCs/Macs matched or beat large cloud LLMs on more than 80% of tested chat/reasoning tasks, used 50%–80% less energy, but still only kept up on about half of the hardest reasoning tasks. 

A Cat’s Commentary

P.S: Before you go… have you subscribed to our YouTube Channel? If not, can you?