RL fails without midtraining: Qwen3-14B stays at chance on FrontierVEP
LoubnaBenAllal1 · x · 2026-11-01
Midtraining is what makes RL work here: RL directly on Qwen3-14B stays at chance (50%) on FrontierVEP, while the same RL recipe on the FineDNA-midtrained model keeps climbing to 0.63 reward/accuracy.
Related event: FineDNA midtraining proves essential for RL on DNA tasks(2 posts)→
More from Research
- If your AI scores 100%, your eval is wrong — an AI safety researcher's eval playbook — AI Engineer · 2026-11-01
- Control Vectors experiment: letting an agent manage its own activations — sloppenheimer · 2026-11-01
- FlashRSI open-sources RSI research: 12 algorithms, 30+ benchmarks, OOD gains still elusive — HuaxiuYaoML · 2026-11-01
- Claude designed scorpion toxin binders: 17 of 211 lab-tested designs bound — shoumikgoswami · 2026-11-01
- Carbon-X: midtrained Qwen3-14B beats every open LLM on DNA benchmarks — LoubnaBenAllal1 · 2026-11-01
- Hugging Face releases FineDNA: 300B+ token dataset to teach LLMs DNA — LoubnaBenAllal1 · 2026-11-01