Brief, mostly signal
Essay asks whether AI chain of thought reflects real computation
An opinion piece arguing reasoning text may not faithfully show internal computation. It cites reports that OpenAI's latest CoT text became hard to read.
What the research found
Opinion essay arguing AI chain of thought may not faithfully reflect internal computation, citing reports that OpenAI latest model CoT text lost word spacing and became hard to read; background only.
Claims and how well the evidence supports them
- OpenAI latest models show spaces removed from chain of thought text 93%
- AI chain of thought may not faithfully reflect internal computation 97%
Evidence
- note.com https://note.com/mackey1966/n/nf0c5a7b69531?hl=en
Actionability rated none.