AI
Knowing When Thinking Is Not Enough: Teaching Small Reasoning Models to Reason Beyond Their Parametric Knowledge
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 27/09/2026, 20:00:00 UTC
Briefing created: 29/09/2026, 08:29:40 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 29/09/2026, 05:18:25 UTC.
Briefings
Research shows self-refinement consolidates probability mass onto existing solutions rather than enabling new ones, revealing two failure regimes: execution bottlenecks and knowledge bottlenecks. The FlyBy framework introduces selective querying to address these issues, training 4B and 8B variants to reason first, diagnose unresolved gaps, and query stronger models when knowledge is missing. On 1,158 hard problems across six benchmarks, FlyBy-4B achieved 45.96% pass@8, outperforming Qwen3-14B at 2.7x lower cost while exceeding Qwen3-8B in pass@1.
Key points
- Self-refinement interventions consolidate probability mass onto solutions already reachable from the current state rather than making new ones reachable.
- FlyBy-4B achieved 45.96% pass@8 on 1,158 hard problems, surpassing Qwen3-14B at 2.7 times lower serving cost while exceeding Qwen3-8B in pass@1.
Context and limitations
Results are limited to 1,158 hard problems across six benchmarks; findings apply to small reasoning models (sRMs) and two model families at multiple scales.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 29/09/2026, 08:29:40 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeibawmsi6x6ncaqkjo4owpbosfwrq4jrdy7bged3dqlcjmbacg6jkm