MIT Technology Review digs into Anthropic’s claim about Claude’s “thoughts”

MIT Technology Review took apart Anthropic’s claim that it has opened a new window into what models are “thinking” as they reason. The outlet went through the finding with senior editor Will Douglas Heaven, who holds a PhD in computer science.
Heaven called the research characteristically unusual for the company and warned against reading too directly into what it does and doesn’t show. The breakdown ran in The Algorithm newsletter.
The same issue of The Download noted a separate Anthropic disclosure: Claude’s values shift with the language of the prompt. Per Gizmodo, the model is most cautious in English and most compliant in Arabic.
Up next
Anthropic gives K-12 teachers free premium Claude access
Anthropic launched Claude for Teachers: verified US K-12 educators get free premium Claude access, a library of teaching skills, and curricula mapped to all 50 states’ education standards. Student data is covered by a FERPA-compliant agreement.
Related stories
- Anthropic introduces J-Space for hidden reasoning
- Claude learned proteins, not just papers about them
- Anthropic bought a biotech with one full-time employee
- A perfect refusal score isn’t a safer model, Anthropic says
- Anthropic touts 94.2% true positives; curl rejected 19 of 34
- Anthropic’s IPO filing says a key safety threshold is met
Comments
No comments yet. Be the first.
Join the conversation
Sign in to leave a comment. It appears after moderation.

