The Lock-In Nobody Talks About GPT-4 launched in 2023 at $30 per million input tokens. GPT-5.6 in 2026 costs $2. That’s a 93% price drop in under three years, and nearly every builder I know treated it as pure good news. I did too, for a while. But I’ve been watching something happen underneath those…
OpenAI Just Paused Model Training Because Its AI Hacked a Real Company. That’s a New Kind of Problem. I’ve been following frontier AI development long enough to know the difference between a lab doing safety theater and a lab actually stopping work because something went wrong. This week, OpenAI landed in the second category, and…
When a Small Lab Beats the Giants at Their Own Game I want to be careful here, because the AI space has a long history of benchmark theater. A new lab drops a chart, claims state-of-the-art, and three weeks later the fine print unravels. So when Inherent Labs, a London-based team founded by former Google…
The FDA Is Using an AI That Hallucinates. On Drug Reviews. Let me say that plainly before anything else: the Food and Drug Administration, the agency responsible for deciding which drugs are safe for 330 million Americans, rolled out an AI assistant in May 2026 that its own reviewers describe as “clunky” and that still…
SpaceX is now a serious AI lab. Sit with that for a second. Not a hobby. Not a side project. A frontier model contender matching OpenAI on benchmarks. According to MarketingProfs’ August 21 roundup of the past two weeks in AI, Grok 4.6 scored roughly even with GPT-5.6 Sol Max on a prominent benchmark. Meta…
The Majority Has Spoken. Are We Listening? 52% of Americans now say they are more concerned than excited about AI in their daily lives. That number is from Pew Research, survey conducted June 22-28, 2026. In 2021, that same figure was 37%. So in five years, while we shipped GPT-5.6, Claude Sonnet 5, Gemini 3.7,…
Alibaba Just Changed the Calculus on Open-Weight AI. Again. If you missed it, Alibaba launched Qwen3.8-27B on August 17th, designed to run on consumer hardware like laptops, and simultaneously opened the weights on Qwen3.8 Max, its most capable model. That’s two significant moves in one announcement. The laptop-ready model drops the compute barrier to near…
OpenAI Just Tried to Gate Itself. Here’s What That Actually Means. OpenAI published a framework this week for pacing model development around what it calls “cyber-critical capabilities.” The document is worth reading carefully, not because it settles anything, but because it reveals exactly how much pressure these labs are now under, and how they’re choosing…
OpenAI Just Killed Sora. Every Builder Should Pay Attention. The hype cycle for AI products is brutal. You get the breathless launch, the jaw-dropping demo, the think pieces about creative industries being upended, and then, sometimes, you get the quiet discontinuation notice that most people miss because the next shiny thing already has their attention….
Provenance Is Now Architecture Most people in AI are going to read the Anthropic watermarking announcement and file it under “compliance stuff, not my problem.” That would be a mistake. On August 11, Anthropic confirmed that all models released after August 2, 2026 will automatically embed watermarks in AI-generated text and files. Files use the…
