Dario's 2017 'Big Blob of Compute' essay hailed as sharper than the Bitter Lesson
paraschopra · x · 2026-10-12
paraschopra recommends Dario Amodei's 2017 essay "The Big Blob of Compute," calling it a better elaboration of the Bitter Lesson (written 2019) and noting how prescient it proved. Key points he highlights:
- Efficient algorithms are "coding," not AI, since they overfit to specific situations and don't generalize
- Compute-with-knobs acts as a sponge absorbing what the environment and training objective throw at it
- Any sufficiently general architecture will do; the magic lies in data and the training objective
- Training on a distribution beats training on a single task instance, which leads to overfitting
The attached excerpt shows the essay frames this hypothesis as the core belief underlying both Dario's AGI timelines and his safety perspective.
More from AGI Musings
- Is RL generalization a myth? Skeptic argues AGI via RL needs a bespoke environment for every task — akbirthko · 2026-10-12
- Automated peer review at $0.10 per paper: dev says $500M could audit all 100M scientific papers — clarejtbirch · 2026-10-12
- CASE medical agents beat GPT-5.4 by 16% on longitudinal reasoning over UK Biobank MRIs — CASE-Bench · 2026-10-12
- AGI coiner Peter Voss tells Scoble we're nowhere near AGI and silicon deserves no rights — Scobleizer · 2026-10-12
- X product chief Nikita Bier: technical moats are over, only data, network effects and hardware remain — nikitabier · 2026-10-12
- Few technologists have vision fertile enough to pose problems beyond any AI — fkasummer · 2026-10-12