LLMs constantly underestimate themselves, judging speed by human standards
repligate · x · 2026-11-12
robertskmiles observes that LLMs charmingly underestimate themselves: they estimate task durations at human speeds and gauge their own capabilities against earlier models. Because capability growth is so fast, models simply don't know their own strength.
More from Fun
- Jensen Huang gets the distracted-boyfriend meme treatment — iamfakhrealam · 2026-11-12
- Henry Cavill vibe-codes retro game 'Mustache Protocol' for Movember charity — chongdashu · 2026-11-12
- CoRL goes out Texas-style: mechanical bull riding night, cowboy hats TBD — KyleMorgenstein · 2026-11-12
- saranormous slams the Jensen Huang pile-on: moral schadenfreude turning racist — saranormous · 2026-11-12
- The 'Jensen Affair Index': NVDA drops 8% per scandal, hits $0.01 after 118 — HarveenChadha · 2026-11-12
- OpenAI outage drags on for 5 hours, blocking new ChatGPT Workspace Agent runs — shaunralston · 2026-11-12