Fetching from the wire…
01
02
03
04
05
06
07
08
Models2026-09-26 · source-backed
Anthropic's science blog, written by physicist Matt von Hippel, says Claude beat the eight-loop record Lance Dixon and Andy Liu set in 2023, using both the bootstrap method and an indirect form-factor approach, delivered "in a single shot, without any scientific oversight." About a week on 96 CPUs, $1,000-$2,000. Dixon checked it. Von Hippel's conclusion is the interesting claim: a lot of frontier physics is blocked on heavy engineering rather than new ideas, and that's work agents can already do.
Each link below shares sources, entities, or timing with this story.
Anthropic ran a de novo binder campaign where Claude researched each target's biology, picked docking sites, installed open-source tools from their public repos itself, and composed 24 workflows with no human making a design decision. Of 1,320 designs synthesized and measured...
The letter to Senators Tim Scott and Elizabeth Warren, dated June 10 and surfacing publicly this week, frames it as model distillation run against Claude at scale (Anthropic). A related claim pegs it at 28.8 million fraudulent exchanges, though that figure is single-sourced an...
Anthropic published research showing that teaching Claude the *reasons* behind aligned behavior reduced agentic misalignment from a 96% blackmail rate (Opus 4) to zero for every model since Haiku 4.5. A "difficult advice" dataset did it in 3M tokens vs. 30-85M for synthetic ap...
The most useful engineering blog post I've read this year dropped today with zero fanfare. Anthropic's engineering team published the actual architecture they use for long-running autonomous coding: a two-agent harness where an initializer agent sets up the project environment...
Two competing models for AI-powered security shipped on the same day. OpenAI launched Codex Security ("Aardvark") — an AI AppSec agent that builds project-specific threat models, then hunts for vulnerabilities and tests them in isolated environments. 30-day beta: 1.2M+ commits...
Two facts sit next to each other and neither cancels the other out. Anthropic published on September 4 that an internal general-purpose research model, roughly comparable to Claude Fable 5.1, formalized Fermat's Last Theorem in Lean over 11 days working largely autonomously. T...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.