bira

Ad­ding Ty­pos Made Haiku’s Ac­cu­racy Go Up

• bira 24 Feb 2026 7:02 UTC

  1 point

  0

  in reply to: Roy Rinberg’s comment on: [Er­ror com­mu­ni­cat­ing with LW2 server]

  Got it. I was indeed thinking to frame this blogpost as more lightheartedly. Would this be better: “We added typos to a benchmark, then Haiku’s scores jumped” so there is no mention of “saturation”. I’m thinking the blogpost is more of “why did Haiku’s score jump” instead of “LLMs are robust to typos”.

• bira 24 Feb 2026 5:56 UTC

  1 point

  0

  in reply to: Roy Rinberg’s comment on: [Er­ror com­mu­ni­cat­ing with LW2 server]

  I just think it’s a good hook, taking a leaf from how news pieces are written

Spooky Col­lu­sion at a Dis­tance with Su­per­ra­tional AI

[GDP­val] Models Could Au­to­mate the U.S. Econ­omy by 2027

[CS 2881r AI Safety] [Week 1] Introduction