All editions.
Every briefing, fully archived. Five to six new ones every week.
285 briefings. Five to six new ones every week.
- EDITION #310
Two Models From the Same Base Scored 1.5% and 95% on the Same Test
OpenAI's GPT-5.6-Cyber found two Chrome zero-days. The number that matters measures willingness, not skill.
- EDITION #309
The AI That “Talks to Dolphins” Can’t Translate a Word — and Still Isn’t Released
DolphinGemma finds patterns in dolphin sound. It doesn’t translate, and 15 months after launch it’s still in development.
- EDITION #308
Two Flagship AI Models, 42 Days Apart. The Real Changes Never Reached a Benchmark Chart.
Anthropic shipped Opus 4.8 just 42 days after 4.7. Same price, same context, same cutoff. What changed was the plumbing.
- EDITION #307
This AI Refuses to Answer Unless It Can Prove It. Even Its $27M Makers Won’t Say “Hallucination-Proof.”
Pramaana Labs raised $27M to prove AI answers in Lean. The honest part is the phrase it refuses to use.
- EDITION #306
Coffee and Beef Share 102 Flavor Compounds. Shrimp and Lemon Share 9.
What AI actually knows about your dinner, what it only claims, and the free 56,498-recipe study you can test tonight.
- EDITION #305
The 2% Experiment That 98% of Customers Saw
Anthropic tested a price change on 2% of new signups but changed the public pricing page for everyone. Reverted within a day.
- EDITION #304
One-Third of Entry-Level Jobs Now Want AI Skills. The Edge Is Managing Agents, Not Using Them.
Harvard Business Review named the role in February: agent manager. The hiring data says that's where the leverage is.
- EDITION #303
Attacks on AI Web Agents Partly Work 86% of the Time. Chrome's New One Has Your Passwords.
Gemini Spark can now drive your real Chrome using saved logins. What Google built to keep it safe — and what the research says.
- EDITION #297
I Asked AI Over 500 Questions in Italy. It Never Forgot a Single One.
Three weeks, one AI, hundreds of questions — and the story of what happened when it started remembering the answers.
- EDITION #296
I Fed Claude a Family Card Game’s Rules. It Came Back a Website, a PDF, and a Cheat Sheet.
A decades-old family rules sheet, turned into a mobile page, a full PDF, and a one-page cheat sheet in one sitting.
- EDITION #295
AI That Simulates Physics and Cause-and-Effect Just Hit $1.45B
A Palo Alto lab raised $310M to build AI that predicts reality itself.
- EDITION #294
An AI Tuned Its Own Engine. The Price Fell 80% the Next Day.
OpenAI let its top model rewrite the code that runs it. A day later, its cheapest model fell to 20 cents per million tokens.
- EDITION #293
Why 2026 Became the Year Companies Started Paying for AI With People
More than 40 companies named AI when they cut jobs in 2026. Three ran the same 3-step script. Here's your exposure check.
- EDITION #292
OpenAI Just Built Its Own Chip — and It Used AI to Design It in 9 Months
OpenAI and Broadcom built Jalapeño, a chip for running AI at gigawatt scale, designed in just 9 months using AI.
- EDITION #291
We Ran One Prompt on 3 of Our Own Paragraphs. It Never Wrote a Word.
One interview-style prompt turns any chatbot into a line editor, not a writer. We ran it on ourselves first.
- EDITION #290
Three Apps Just Moved Into Your Search Bar
Canva, Instacart and YouTube Music now plug into AI Mode. Some steps finish there; most still hand off to the app.
- EDITION #289
AI Agents Can Now Finish 15.8% of Real Freelance Jobs — Up From 2.5% Eight Months Ago
240 real freelance jobs, $144,000 on the line, human-graded. What splits the 15.8% that pass from the 84% that fail.
- EDITION #288
A Computer Simulation Designed This Vaccine's Active Ingredient — and It Just Passed a 39-Person Safety Trial
Cambridge's computer-designed coronavirus vaccine is safe in its first human trial. The immune response was modest.
- EDITION #287
40 Companies Just Agreed on How AI Agents Are Allowed to Pay
Visa, Mastercard, Stripe, Google, AWS, Shopify, and Cloudflare just backed the open standard for how AI agents pay online.
- EDITION #286
Why "Turn On Thinking" Is Now an Error Message
The thinking toggle is gone. Claude and ChatGPT replaced it with the same five settings, then picked opposite defaults.
- EDITION #285
Why This Free AI Model Should Worry Every AI Company
One trick called "mixture of experts" explains why a 2.8-trillion-parameter AI model can run for free.
- EDITION #284
Why Your ChatGPT Agent Now Only Needs Your Password Once
ChatGPT Work agents can now log into sites for you — and remember it, so you never repeat that step again.
- EDITION #283
Scammers Deepfaked Ronaldo 254 Times. We Built a Prompt to Catch It.
AI deepfakes are selling fans on illegal casinos this World Cup. Here's a prompt that catches the next fake.
- EDITION #282
12,637 Real AI Conversations Later, Here's What People Actually Ask It For
A new study of real AI conversations found its #1 use, two years running, has nothing to do with spreadsheets.
