AI SAFETY TWITTER UPDATES

← Front page

Post quotes early internal snapshot of Claude Mythos Preview rewriting code; Zvi responds

2 people · 4 posts · first post · latest

A quoted post describes trying an early internal snapshot of Anthropic's Claude Mythos Preview, which reportedly rewrote code; @thezvi responds jokingly, "Hand-writing code? In this economy?" The claim comes only from the quoted post, not from Anthropic. In the same thread, @natolambert released a GitHub repo tracking HuggingFace models for open model analytics and, when asked, replied that one thing in the thread is established fact and another is unknown.

All posts · 4, newest first

  1. @thezvi commentary ·
    I think you have to discount somewhat, but largely no
    2 likes · 0 reposts
  2. @natolambert research ·
    One is established as fact one is unknown.
    19 likes · 0 reposts
  3. @thezvi commentary · · replying to @andrewcurran_
    Hand-writing code? In this economy?
    53 likes · 1 reposts
  4. @natolambert research ·
    Our model tracking repo is here and we welcome corrections! github.com/Interconnects-…
    7 likes · 0 reposts