If you ask a frontier LLM a multi-hop reasoning question, e.g.,...
Kaley Brauer 💫@kaleybrauer
11 views
Aug 16, 2026
~1 min read
Advertisement
1
If you ask a frontier LLM a multi-hop reasoning question, e.g., "Who won the Nobel Prize for Chemistry in (1900 + Mozart's age when he died)?", it usually can't answer correctly immediately (no thinking)
BUT if you ask the same question & append 300 dots, suddenly it can answer?
BUT if you ask the same question & append 300 dots, suddenly it can answer?
2
Clearly the models are using the dot tokens to do extra hidden computation (looking up facts, doing math)
We investigated how + what the models are doing and showed the computation is readable from the hidden states over the "meaningless" filler tokens: arxiv.org/abs/2607.03502
We investigated how + what the models are doing and showed the computation is readable from the hidden states over the "meaningless" filler tokens: arxiv.org/abs/2607.03502
3
Just presented this at #ICML 2026 Mechanistic Interpretability Workshop in Seoul! So many wonderful people doing fascinating work
4
Check out @RyanGreenblatt’s post that inspired this!
blog.redwoodresearch.org/p/recent-llms-…
blog.redwoodresearch.org/p/recent-llms-…
