Extra hidden computations in LLM using dot tokens for multi-hop reasoning

Researchers discovered that large language models can perform hidden multi-hop reasoning by inserting dot tokens into prompts, revealing an emergent computational behavior not previously documented.

Article URL: https://xcancel.com/kaleybrauer/status/2078185882926846044 Comments URL: https://news.ycombinator.com/item?id=48954687 Points: 1 Comments: 0