finding
active
finding:gemini-3-1-pro-auroc-drops-to-0-654-with-nla-and-0-746-with-sae-versus-0-793-baselineGemini-3.1-Pro AUROC drops to 0.654 with NLA and 0.746 with SAE versus 0.793 baseline
Significant downlift for Gemini.
Source paper
extracted_from(2026) · Adam Karvonen · Euan Ong · Subhash Kantamneni · Samuel Marks
Neighborhood — ranked by edge-count
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.
- Second-highest lift; Gemini Pro is the highest-gated model in the study
- Robustness to hyperparameters on Gemma.
- G2.5-FL has extremely low points per coin spent, indicating poor resource use.
- Soft clues redundant with transcript.
- G3.1-FL frequently exercises the buy-right to keep auctioned cards.
- Emergent scaling trend showing VS better exploits capabilities of larger models
- Baseline comparison from prior work used to contextualize insecure variant S values
- G2.5-FL shows no phase adaptation in bidding intensity.