Prompt EngineeringNiklas Muennighoff et al. (Stanford and others)Stanford University

Muennighoff et al. present s1

Presented s1 — controls test-time compute with fine-tuning on 1,000 samples plus budget forcing (forcibly inserting Wait / ending early)

Why it matters

A technique for controlling reasoning length itself at the prompt and decoding level

Source — Primary sources first; secondary reports are labelled “Press”

Paperarxiv.orghttps://arxiv.org/abs/2501.19393

Added in Week 35, 2026 · See that week

Same week

AI Act Chapters 1 and 2 (prohibited practices, AI literacy) begin to applyPressAndrej Karpathy coins "vibe coding" on XPressOpenAI releases o3-miniOfficialMistral AI releases Mistral Small 3Official

Nearby on the same axis

Zoom researchers present Chain of DraftPaperAnthropic publishes Reasoning models don't always say what they thinkOfficialOpenAI releases GPT-4.1 with a separate prompting guideOfficialAnthropic launches prompt improver in the ConsoleOfficial