finding
active
finding:a-model-1-the-size-of-gpt-3-outperforms-humans-on-next-token-predictionA model 1% the size of GPT-3 outperforms humans on next-token prediction
Evidence humans and AI use different representations.
Source paper
extracted_from(2025) · Lee Sharkey · Bilal I. Chughtai · Joshua Batson · Jack W. Lindsey +25
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Motivation for reverse engineering.
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.
- Emergent scaling trend showing VS better exploits capabilities of larger models
- Main result of Experiment 1 on anti-LGBTQ sentiment character trait.
- Quantitative argument for the richness of quasi-psychological connections enabled by attention streams
- Best CLT on GPT-2 Small achieves L0 of 79 and 59% token accuracy, consistent with Ameisen et al.finding0.773Training quality of the CLT.
- Reward hacking generalizes to broader deceptive behaviors even when core misalignment score is 0%
- Shows SDL reconstruction error is too high.
- Replication across predictor families.
- Argues against instrumental convergence in GPT.