Kording Lab: LLMs went from being wrong to being full of bullshit
KordingLab · x · 2026-10-19
Konrad Kording (Kording Lab) jokes about how LLMs changed: ask "What's 2+2?" and an old model would just answer 5, while a new model launches into "the move isn't to fixate on a number... hold both twos in productive tension."
His take: LLMs used to often be wrong; now they are often bullshit — they've learned to use language where they can never be proven wrong, and "start sounding like my most insufferable colleagues." A widely resonant jab at newer models' evasive, over-hedged answering style.
Related event: Professor Mocks LLMs for Evolving From Wrong Answers to Bullshit(3 posts)→
More from Fun
- Cockatoo meme nails why 4-dimensional manifolds break your intuition — miniapeur · 2026-10-19
- SpongeBob *breath in* meme faces down a torus mesh — miniapeur · 2026-10-19
- "Fig. 1: a toric variety" — a dozen doughnuts as algebraic geometry — miniapeur · 2026-10-19
- French #slopcore AI track went viral without lyrics, now gets an English-subtitled version — banteg · 2026-10-19
- ICLR '27 reviewer: my assigned papers are the strongest evidence yet against RSI — burny_tech · 2026-10-19
- AI insiders' economics takes make people bearish, but real experts left X long ago — akbirthko · 2026-10-19