hypothesis
active
hypothesis:pretrained-llms-may-show-fewer-signs-of-post-hoc-reasoning-than-rlhf-models-e-g-changing-their-answer-more-when-mistakes-are-addedPretrained LLMs may show fewer signs of post-hoc reasoning than RLHF models, e.g. changing their answer more when mistakes are added.
Future work in limitations.
Source paper
extracted_from(2023) · Tamera Lanham · Anna Chen · Ansh Radhakrishnan · Benoit Steiner +26
Neighborhood — ranked by edge-count
Papers (1)
paper
Questions (1)
question
- Are training schemes different from RLHF more effective at eliciting faithful reasoning from LLMs?associated_withFuture work.
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- Binder et al. finding cited as evidence that LLMs possess introspective capacity analogous to mindfulness
- Opening framing.
- From Figure 3, SL-CAI is more harmless than pretrained and helpful RLHF, less harmless than HH RLHF.
- Establishes that the observed linear structure is not merely a representation of text probability
- Theoretical hypothesis about the mechanism underlying LLM error detection and reflection.
- Out-of-context reasoning work directly related to synthetic document fine-tuning experiments
- Claim that capability emerges from architecture, not data, and that later models lose the surprise.
- Supporting evidence for cross-modal platonic representation