TAWS

From the Friday, November 6 show

METR estimates Anthropic is improving its AI about twice as fast

AI-generated from the transcript

What happened
Watchdog group METR spent eleven weeks reviewing Anthropic's internal data and estimates Anthropic now improves its AI about twice as fast as in 2024 to 2025, guessing AI drives most of that. In the public report, Anthropic blacked out most of METR's key evidence. METR checked Claude's work against staff interviews and found no clear bias, but most of its analysis relied on Claude.
Austin's take
Austin flags a possible conflict of interest, saying many METR founders were early Anthropic investors. He calls this RSI Lite, expects documented recursive self improvement around early 2027, and is excited, while saying safety findings should be shared even if competitive details stay private.
Why it matters
If AI is speeding up its own development, keeping humans able to understand and audit the process gets harder fast.
What to do
Read claims about AI speedups with an eye on who did the checking and what was redacted.
Not confirmed yet
METR's figure is an estimate, its attribution to AI is a guess, key evidence was redacted, and the review leaned heavily on Claude itself.

Watch from 22:22 on the episode page

The Loop

The weekly newsletter

The week in AI in one email: what happened, why it matters, what to do about it.