Data scientist: AI's good results come from expert human context, not model magic
gerardsans · x · 2026-10-14
Data scientist Gerard Sans argues AI only gets good results when a human supplies enough context to narrow the target direction — the 'million dollar prompt' takes years of expertise, and AI labs never disclose how much humans actually contribute to the success. In a follow-up he calls current AI 'wasteful, blind and stupid': point it the wrong way and it won't tell you the target is behind you. A pointed counter to AI capability hype.
Related event: Developer Argues AI Hitting the Target Doesn't Make It a Sharpshooter(3 posts)→
More from AGI Musings
- Mecha-nudges: subtle environmental tweaks that steer AI agents but not humans — ethayarajh · 2026-10-15
- Goodfire's interpretability system EVEE reclassifies hearing-loss gene variant as likely pathogenic — fatihdin4en · 2026-10-15
- The small-model world has changed forever, but the builder community hasn't priced it in — tokenbender · 2026-10-15
- Ziz cult recruit recalls being offered free 'mind-fixing' in exchange for psychological info — jessi_cata · 2026-10-15
- Why hasn't Anthropic had an OpenAI-style misalignment incident? Three possible explanations — gleech · 2026-10-15
- SF Is Full of 26-Year-Old Deca-Millionaires: A Founder's Guide to Not Chasing the 0.001% — brandon_galang · 2026-10-15