Claude guessed LLMs won't write full PRs until 2028 — even models can't grasp their own pace
felixrieseberg · x · 2026-10-29
Developer felixrieseberg notes how hard it is to internalize that models improve at an accelerating pace — even people at Anthropic and OpenAI struggle, and so does Claude itself.
He asked different Claude generations the same questions:
- Opus 3 guessed LLMs won't reliably write entire PRs without human guidance until 2028-2030;
- Opus 5.5, asked about the exponential time hypothesis (ETH), said a resolution likely comes after 2045 — noting proving ETH is harder than P≠NP, while disproving it (a sub-exponential 3-SAT algorithm) is the kind of thing AlphaEvolve-style LLM search might stumble onto (5% chance).
The punchline: "little baby Claude has no idea how quickly it's growing up" — old models' conservative forecasts keep getting outrun by their successors.
More from AGI Musings
- Altman's 'don't do the deferred life plan' advice strikes a chord about spending your 20s — menhguin · 2026-10-29
- Dev criticizes 'Tall Poppy' mindset after publishing AI research within months of starting — menhguin · 2026-10-29
- davidad: STEM hobbyists should let go of the frontier, not defend it — davidad · 2026-10-29
- Scott Aaronson: AI labs are solving century-old math problems to show off before IPOs — AnnaCiaunica · 2026-10-29
- Critic says closed weights hide personas from users, but that doesn't prove models aren't general-purpose actors — aiamblichus · 2026-10-29
- Generative AI is adversarially optimized to seem better than it is, warns Ivan Vendrov — dhadfieldmenell · 2026-10-29