Bitter lesson part 2: few academic inventions matter, and conference publications aren't the key
srchvrs · x · 2026-10-27
srchvrs argues only a small fraction of academic inventions make a difference — and they aren't always published (Muon optimizer) or at prestigious venues. Slides list what LLM developers actually pick up: solves an important problem, strong continued software support, strong baselines, demonstrated scaling; conference pubs seem less important (4/5 were at good conferences, Muon wasn't). Broader audience criteria: strong usability, beats the next thing, works in real life. He cites GitHub crediting only one of three authors of their fast bit-packing work.
Related event: 'Bitter lesson 2': few academic inventions matter, venue is irrelevant(2 posts)→
More from Research
- Anthropic x Adaptyv protein design challenge targets mutant KRAS for cancer immunotherapy — nc_frey · 2026-10-27
- Prima Mente's Pleiades-Neuron detects neuron-derived cfDNA in blood, 4.4x more sensitive — iScienceLuvr · 2026-10-27
- MIT: LM harnesses are compositional generalizers, not just scaffolding — akbirthko · 2026-10-27
- InterviewPlayground: JHU's simulated eval for AI interviewers hits 0.86 correlation — anjalie_f · 2026-10-27
- Formal methods expert: never trust LLM-generated specs, vet them instead — davidmanheim · 2026-10-27
- Reasoning models can barely control their chain of thought, paper finds — tomekkorbak · 2026-10-27