Sharpening Tax: Does post-training teach LLMs new agentic skills or just sharpen old ones?
SharonYixuanLi · x · 2026-10-02
Researchers led by Changdae Oh introduce "Sharpening Tax," a framework probing whether post-training actually teaches LLMs new agentic skills or merely sharpens capabilities the model already has.
The work aims to measure the hidden cost of post-training — improvements to existing abilities that may come at the expense of genuinely new skills. Full data and experiments are in the original 11-post thread.
More from Research
- KL divergence between Cauchy distributions is finite and symmetric, with closed form — FrnkNlsn · 2026-10-02
- Learner to share daily notes through Stanford's CS224n NLP course — stanfordnlp · 2026-10-02
- Building CS224n's Transformer from scratch: 144 lines expose scaling and mask bugs — stanfordnlp · 2026-10-02
- One human health check erases semaglutide's behavioral effect in mice, preprint finds — BenSiranosian · 2026-10-02
- BeyondSCe grasps objects by past-event reference, hitting 77% success on occluded targets — SeoulNationalUniv · 2026-10-02
- RouteFM pretrains LLM routing once for anywhere: +2.23 quality points over strongest baseline — nanjinguniv · 2026-10-02