Why RL Struggles With Fiction: No Ground Truth, Unlike Code, Math and Lean
teortaxesTex · x · 2026-09-22
Responding to the question of why AI writes poor literary fiction, teortaxesTex notes targeted attempts exist but it's not a priority and is genuinely hard: there's no ground truth. "How do you even score online? With LLM judges?" Coding has tests, calculation has instant checking, Lean makes math verifiable — fiction lacks any such verifiability, which is why RL hasn't cracked it.
Related event: Why AI Can't Write Serious Literature: Taste Variance vs Artistic Intent(6 posts)→
More from Models
- Grok 4.7 launches on Cursor and API, topping coding benchmarks at half the price — FinanceYF5 · 2026-09-22
- Grok 4.7 launches at same pricing: Terminal-Bench doubles to 38%, 500K context kept — FinanceYF5 · 2026-09-22
- Claude Status: Elevated Errors Reported for Multiple Models — corvad · 2026-09-22
- tenobrus and antirez pour cold water on Jev: demos are inflated and far from functional — burny_tech · 2026-09-22
- Musk confirms Grok went from outside top 10 to top 3 in 90 days, Grok 4.8 next — elonmusk · 2026-09-22
- Anthropic investigates elevated errors across Claude Mythos 5.1, Fable 5.1 and Opus 5 — ClaudeAI-mod-bot · 2026-09-22