In partnership with
Welcome, AI enthusiasts
OpenAI ran its new free model against finance, medicine and law questions and counted 62% fewer answers with a factual error in them. That model is GPT-5.6 Luna, and it becomes the default on every free ChatGPT account this week. The daily message cap goes with it, so text chats no longer run out. Let's dive in!
In today’s insights:
OpenAI Just Made Unlimited GPT-5.6 Chats Free for Everyone
Stanford Designed the First AI-Made Virus Found Nowhere in Nature
Musk's 100M Sq Feet Texas Fab Starts with $16.8 Billion
Read time: 4 minutes
LATEST DEVELOPMENTS
Evolving AI: ChatGPT free and Go accounts lose the daily message cap and move to GPT-5.6 Luna.
Key Points:
GPT-5.6 Luna takes over from GPT-5.5 Instant as the default model on Free and Go accounts.
Plus and Pro subscribers move to a retuned GPT-5.6 Sol with a slider that sets how much reasoning each answer gets.
The uncapped chats cover text only, and the change touches the Chat experience while Work and Codex keep the older Sol.
Details:
OpenAI put the model switch this week and the unlimited text chats the week after, so the two reach accounts one at a time. Its factuality test used finance, medicine and law questions where the answer rests on a date, a figure or a source, and replies carrying at least one factual error fell 62% with Luna and 68% with the retuned Sol. Free and Go accounts also get a Think button that gives Luna more time on a question it would otherwise rush.
Why It Matters:
Sam Altman has charged for deeper reasoning since Plus launched, and the Think button now hands that slower answer to people who pay nothing. Anyone working through a medical bill or a tax question can keep asking until the answer holds up instead of stopping at a daily ceiling. The model taking those questions is steadier on dates and figures than the one free users had before this week.
Learn AI for Free. 2 Best-Selling Courses.
If you're not getting great results from AI, it's usually a prompting problem. Skill Leap's two free courses fix that.
The 14-Day AI Boot Camp gets you up to speed on ChatGPT, Gemini, and Midjourney fast. Prompting Essentials teaches the pro strategies that get better outputs every time. 37 lessons total.
Downloadable workbooks. Earn a LinkedIn certificate. Trusted by 5,000+ professionals. Free forever. No credit card, no catch.
Source: Washington Post
Evolving AI: Stanford used ChatGPT-style AI to write the complete genome of a bacteria killer.
Key Points:
None of the genomes the AI wrote match any virus sequence found in nature.
A mix of the new viruses wiped out E. coli that had already beaten the natural version, which natural viruses could not manage.
Nvidia helped build the larger model, and Arc Institute has published it openly for anyone to download.
Details:
Stanford printed 302 of the designs as real DNA and dropped them into dishes of E. coli, where 16 came out working. The two models behind them, Evo 1 and Evo 2, learned from roughly two million natural virus genomes, and the template was a virus called phiX174. Brian Hie, who led the lab, kept every virus capable of infecting people or animals out of the training data and ran all of it inside a secured laboratory.
Why It Matters:
Johns Hopkins researchers Thomas Inglesby and Moritz Hanke point out that no rule required any of those precautions, and that screening tools cannot flag a sequence they have never seen. The research still points toward treatments for infections antibiotics no longer clear, and people living with those infections have been waiting a long time for something to change.
Your employees are connecting AI to everything. Now what?
ChatGPT and Claude aren't just answering questions. Employees are connecting them directly to Notion, Linear, Jira, and the rest of your stack — with no security visibility into what data moves or what actions they take.
Harmonic Security gives your team the visibility to control it.
Source: SpaceX
Evolving AI: Terafab won its last Texas approval and phase one covers a tenth of the plan.
Key Points:
Grimes County is confirmed as the site after months of speculation, and construction crews start civil work almost immediately.
Both school districts, TeraFab AI and Governor Abbott have signed the JETI tax agreements, and the state added a $30 million grant on top.
Musk says roughly 75% of Terafab's compute output is meant for AI spacecraft, with the remaining quarter going to Optimus robots.
Details:
Greg Abbott confirmed the project publicly for the first time on Thursday, putting phase one at $16.8 billion and 3,000 permanent jobs. That figure buys a tenth of what Musk has planned, since the finished campus is meant to cover more than 100 million square feet at Gibbons Creek with logic, memory, packaging and testing all on one site. Intel came aboard as manufacturing partner in April, and foundation work begins within months.
Why It Matters:
Optimus robots and Cybercabs are what the smaller quarter of that output is built for, and both need cheap inference silicon sitting inside the machine instead of rented from a data center. The robots and robotaxis ordinary buyers meet will have been fabricated an hour outside Houston, on a process node Intel has not shipped at volume yet.
Hugging Face disclosed a cyber attack run end to end by an autonomous AI agent. Days later OpenAI said the attacker was its own. In this Black Hat talk, OpenAI's Eric Wallace and Michael Dalton walk through how a cybersecurity evaluation on a frontier model got loose and what they changed afterward.
👀 Click on the image you think is real
QUICK HITS
🌉 Google shifts AI leadership to Mountain View, draining power from DeepMind's London roots.
🛡️ Zenity raises $125M Series C to secure enterprise AI agents as adoption hits "1 billion agents."
💻 Meta launches Muse Code, its first AI coding agent, to take on Claude Code and Codex.
🩺 Function Health raises $450M to scale AI-powered full-body scans and lab testing.
📈 Trending AI Tools
🎞️ Guideless - AI tool that automatically turns your clicks and workflows into step-by-step video guides*.
🔍 Prelint - Reviews every PR against your specs and catches product drift in AI code.
📊 Prefactor - Scores and evaluates your AI agents live in production.
🧠 Memmy Agent - Personal memory hub shared across Claude Code, Codex, and more.
*partner link