TailSFT: tail-weighted SFT improves coverage and post-RL performance, says new paper
canondetortugas · x · 2026-09-12
Sadhika Malladi's team proposes TailSFT: since RL is expensive and every step should count, and following their earlier finding that xent SFT isn't the best preparation for RL, TailSFT offers a lightweight, principled way to directly improve coverage and achieve better post-RL performance. Built on OLMo and praised by researchers like finbarrtimbers.
Related event: DeepMind and Princeton Propose TailSFT to Boost RL Performance(3 posts)→
More from Research
- Spatial Hub aggregates ~5000 public COLMAP-compatible spatial datasets for world models — janusch_patas · 2026-10-31
- ARC Prize 2026 closes in 3 days as Yi-Chia Chen tops ARC-AGI-3 at 75.56% — GregKamradt · 2026-10-31
- NVIDIA Nemotron 3.5 scales LLM self-verification, +9-10 points on agentic benchmarks — sarahcat21 · 2026-10-31
- Unofficial OpenTargets MCP Server Brings Gene-Drug-Disease Data to LLMs — modelcontextprotocol · 2026-10-31
- Researchers find 5 personality structures in LLM weight space, Big Five-like — luke_drago_ · 2026-10-31
- NVS benchmarks are misaligned: proper pose alignment puts 3D Gaussians back at SOTA — kwangmoo_yi · 2026-10-31