Google AI's chain of thought on Ulam sequences is brilliant and dumb at once
OmegaCookieMonster · reddit · 2026-11-01
A Reddit user shared screenshots of Google AI's visible chain of thought while deriving the Ulam sequence (each next term is the smallest integer uniquely representable as a sum of two distinct earlier terms).
- For U6 the model does well: it shows a very human "wait" moment, writes down 8, then carefully checks 2+6 and rules out 1+7 (7 not in sequence) and 3+5 (5 not in sequence) to confirm uniqueness.
- For U7 it flops: even though 7 obviously isn't in the sequence, the model goes off to check OEIS's standard sequence, prompting the poster's rant that it "clearly has intelligence, but can also be such a dumbass."
The poster admits it's a minor example, but it captures the classic vibe of LLM reasoning: capability and silliness coexisting.
More from Fun
- Halloween meme: the scariest thing in your kid's candy bucket is a Teams install disc — flngr · 2026-11-01
- Joke: using $5 of Palantir API credits to find out who unfollowed me — flngr · 2026-11-01
- Karpathy vibe-codes HackerTube in 2 hours with Opus 5.5: 6,454 HN-recommended videos — daniel_mac8 · 2026-11-01
- Karpathy vibe-built HackerTube archives decades of HN-recommended YouTube videos — daniel_mac8 · 2026-11-01
- Fan dresses up as Claude Code to get a commit — and still hits the weekly usage limit — tech__unicorn · 2026-11-01
- 'Your usage will reset after 5 hours' — the exact face you make waiting for your AI quota — Gauri_the_great · 2026-11-01