Reading up on natural-language-processing
2 deep · digging since mar 12
- Don't let the LLM speak, just probe it. - by James Padolsey
By extracting an LLM's hidden state at the last prompt token and training a small MLP, you get a fast, calibrated classifier for any English criterion.
- [2603.09906] Thinking to Recall: How Reasoning Unlocks Parametric Knowledge in LLMs
Reasoning helps LLMs recall simple factual knowledge via a computational buffer effect and factual priming, but hallucinated intermediate facts increase final errors.