Self-explanation training generalizes beyond narrow hint formats to held-out evals
a_karvonen · x · 2026-09-05
Prior self-explanation training generalized narrowly (e.g., one hint format to another) due to narrow datasets. The authors' training never targeted hints, yet both models improve on the held-out hint eval ("would removing the hint change your answer?") and on other evals.
More from Research
- Three reasons AI is not conscious and may never be, per Nature essay — anilkseth · 2026-10-21
- Aviation tracks every part's provenance — agents can reason over that graph — colinmcnamara · 2026-10-21
- SFT details: instruction/reasoning/code/math mix, synthetic + human traces, no epoching — TimDarcet · 2026-10-21
- Mid-training recipe disclosed: 32k→131k context extension, dropping logit distillation for tokens — TimDarcet · 2026-10-21
- Mathematician Daniel Litt concedes his AI bet, but says AI won't solve math — ziv_ravid · 2026-10-21
- AI drug discovery reality check: 'give me everything that binds,' then filter for toxicity — burny_tech · 2026-10-21