Ethan Mollick: GPT-4 could work solo for 4 minutes; today's models run for weeks
eldonredwards · x · 2026-11-02
Ethan Mollick cites the 50%-success task-length horizon metric: GPT-4 could handle 4 minutes of independent work, while current models can work autonomously for multiple weeks. He argues January 2026 is a reasonable cut-off for when agentic work began — large productivity gains before that will mostly be confined to coding, and the chatbot era was never going to deliver a big productivity boost. Quoted poster Alex Olegimas agrees analyses should set the line at Jan 2026.
More from AGI Musings
- In an age of zero marginal cost, attention and personal brand are the only scarcity left — eter_inquirer · 2026-11-02
- Prediction: we're nearing peak ESL as free AI translation erodes English's career value — intellectronica · 2026-11-02
- Michael Levin's Platonic theory: minds as discoverable patterns, AI as interface — Forsaken-Promise-269 · 2026-11-02
- 'Knowing AI' takes 20 seconds to learn — what matters is what you say to it — feregri_no · 2026-11-02
- Scobleizer: Talking to AGI is far slower than thinking to it — new interfaces will unlock superhuman abilities — Scobleizer · 2026-11-02
- 45% of PM job seekers now get take-home homework, and AI is changing grading — aakashgupta · 2026-11-02