Training models to fill in omitted reasoning from proprietary model outputs
OwariDa · x · 2026-11-23
OwariDa proposes training models specifically for fill-in-the-middle tasks that reconstruct omitted reasoning from proprietary model outputs.
Design notes:
- The reward function should be based on whether prefix + filled-in reasoning leads the target model to the right answer (or best next step) in a separate pass;
- The dataset must consist of problems the target model cannot already solve, otherwise the reconstruction has no value.
More from Research
- Anthropic researcher alpoge reportedly solved one of the biggest open math problems — and went silent — kimmonismus · 2026-11-23
- Princeton study: diffusion models barely extrapolate basic visual skills, ~50% on counts — NielsRogge · 2026-11-23
- MIT Professor: Models Cross a Discovery Threshold Around 0.76 Log-Perplexity — ProfBuehlerMIT · 2026-11-23
- Noam Brown's 2023 bet lands: 10K agents solve Navier–Stokes, Riemann proof claimed — haider1 · 2026-11-23
- NeurIPS paper dissects the Platonic Representation Hypothesis and proposes pair-free NGMA alignment — loremstasum · 2026-11-23
- Riemann proof drama: Lean community certifies it, but a geometry predicate defined as constant True raises doubts — rickasaurus · 2026-11-23