TAWS

From the Tuesday, November 3 show

Anthropic warns AI may soon speed up its own progress

AI-generated from the transcript

What happened
In an updated stock filing released Monday morning, Anthropic warned that future models may soon cross a safety line the company drew itself, tied to AI accelerating its own development. No current model has crossed that line. The filing says Anthropic has missed big jumps in its own progress until after the fact. It also shows $17 billion in sales over three months, more than the entire first half of the year, and a $567 million operating loss mostly from compute costs.
The claim
Anthropic says AI may soon be doing much of the model-training work that used to require human researchers, but that no current model has crossed its safety threshold.
Austin's take
Austin frames this as recursive self improvement (RSI) and calls it an exciting step forward. The catch is verification: humans checking AI-built work is slow, which can undercut the speed gains. He argues that if RSI is already here, political efforts to slow AI (like fighting data centers) may barely matter. He also notes a stock filing is partly a sales pitch.
Why it matters
If AI starts doing most of the work of building the next AI, progress could outrun the people responsible for checking it, and the company itself admits it has missed big jumps before.
What to do
Watch how labs describe their safety thresholds and human oversight in future filings and releases. Treat financial disclosures as both information and marketing.
Not confirmed yet
Austin mentions rumors that other labs, including Google, have already achieved recursive self improvement, based on Google's fast jump to a top model. These are unconfirmed rumors.

Watch from 1:55 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.