Context Widows - by Kevin Baker - Artificial Bureaucracy
artificialbureaucracy.substack.com · 3,112 words · saved by 2 readers
or, of GPUs, LPUs, and Goal Displacement
The argument about whether large language models can “do science” has the structure of a trap. The debate proceeds along predictable lines: enthusiasts point to protein folding, to materials science, to the generation of novel hypotheses; critics point to hallucination, to the absence of understanding, to the difference between pattern-matching and insight. But answer yes and you’re a booster, answer no and you’re a skeptic, and either way you’ve made the technology the protagonist. Both sides assume the only question worth asking is a capabilities question: what can these systems really…
saved by
related reading
- The machines are fine. I'm worried about us.ergosphere.blog
- A Vision of Metasciencescienceplusplus.org
- The Need for Verification in AI-Driven Scientific Discoveryalphaxiv.org
- A Vision of Metasciencescienceplusplus.org
- Cargo Cult Sciencecalteches.library.caltech.edu
- An invitation to a secret society - by Adam Mastroianniexperimental-history.com
- Designing AI for Disruptive Scienceasimov.press
- Science as Collective Sensemakingrepublicofscience.substack.com
- The Legibility Problempress.asimov.com
- Could AI slow science?normaltech.ai
- Tech | LILAlila.ai
- Constellations of Borrowed Light · Velaborrowedlight.org