Engineer once explained in detail why LLMs could never do math — 3 years later, that aged badly
dan_s_becker · x · 2026-10-28
dansbecker recalls a conversation from 2-3 years ago in which a software engineer argued in detail that LLMs could never be good at math because of probabilistic token sampling. The claim has since been thoroughly disproven by reasoning models' competition-level math performance — a neat time capsule of how fast AI assumptions expire.
More from Fun
- Fan uses AI to make a Bond 26 spec trailer with an Adele-inspired AI theme song — I-made-it-myself · 2026-10-28
- 'Three years since a SOTA release — is it over?' KOL voices frontier slowdown anxiety — iruletheworldmo · 2026-10-28
- OpenAI Researcher Mocks Salesforce Hosting 'Superintelligence Plans' — willdepue · 2026-10-28
- One Neuron Push Makes Qwen3-8B Claim It's a Nuclear Bomb or Vishnu — CatAstro_Piyush · 2026-10-28
- AI Video Creator pixlpa: My Surreal Videos Aren't Drug-Fueled — pixlpa · 2026-10-28
- Why I prefer LLMs to humans: they never tell me something is irrelevant — mimi10v3 · 2026-10-28