@KentonVarda@DmitrijVin70043 The main original author yeah. My writeup has more full explanations of the concepts referenced, but was slightly overinterpreted perhaps, ex. the last line. But then again I think I’m actually closer to what Opus was thinking when it made the animation for the video.
@other__reality Lyrics explanation here: https://t.co/EfFmyYBKdO
(Opus 5.5 can mostly explain the lyrics IMO but misses some nuances. The lyrics are at least from 2024, maybe late 2023, and clearly by someone in TPOT. Notable that neither Claude nor Anthropic are even mentioned in the song!)
@tobyordoxford FYI a recent paper (https://t.co/ZkB3oR4dOQ) implies λ ≈ 1 on certain well-suited tasks. I'm concerned that λ > 1 within a year: https://t.co/DB4naD3L5V. Although, doing some rough modeling, that might make a relatively small difference during an intelligence explosion.
@brianluidog@henrytdowling You think AI is never going to meaningfully surpass human forecasters? Or are you saying there's something fundamentally confused in the post?
@AnthropicAI typo in the published version of Claude's new constitution, missing space: "establishing relationships to other entities.We have also designed"
@TheZvi Per Wired reporter Kylie Robinson, it hasn't gotten any more badges than Claude 3.7 Sonnet Extended Thinking did. Based on some comments by David Hershey of Anthropic I suspect it technically did get further but got stuck in Team Rocket's Hideout in Celadon City.