Research has cut LLM costs over 10x, and model architecture is the only math lever, argues thread
ChengleiSi · x · 2026-09-10
A widely shared thread explains why labs like DeepSeek and Kimi keep pushing efficiency research: once LLMs handle most everyday tasks, buyers will pick cheaper, privacy-respecting models. Research has already cut costs 10x versus a year ago, and excluding hardware, model architecture is the only factor that mathematically determines cost per million tokens. Ideas, unlike compute, cost nothing to generate.
More from AGI Musings
- The simple accountability rule: AI labs should be fully liable for problems their systems cause — gerardsans · 2026-09-10
- Bryan Johnson responds to Michael Levin's peer-reviewed Platonic Space paper: bodies as collective intelligence — AllThingsApx · 2026-09-10
- Mathematicians Push Back Against AI Lab's 'Mathathon' Compute-Heavy Paper Scooping — _lewtun · 2026-09-10
- Professor: training a PhD takes 5 years, but AI iterates models every few months — DimitrisPapail · 2026-09-10
- Jack Clark proposes pre-registering AI economy forecasts to score predictors in a year — jackclarkSF · 2026-09-10
- Jack Clark: cross-walk the economy in a year against AI scenario communities' predictions — jackclarkSF · 2026-09-10