Floridi et al. Paper: LLM Reasoning Is Stochastic, Not Real Abductive Inference
Grady_Booch · x · 2026-11-23
Grady Booch, arguing with Anders Sandberg and cgcardona that maintaining constraints is fundamentally different from evolving theories, cites Floridi et al.'s new arXiv paper (2512.10080) backing his claim that LLMs are architecturally incapable of abductive reasoning.
The paper's core arguments:
- Token-completion LLMs generate text from learned patterns, not genuine abductive reasoning
- Outputs only appear abductive because training corpora contain human reasoning structures
- Models can produce plausible ideas and mimic commonsense reasoning without grounding in truth, semantics, verification, or understanding
- This dual nature — stochastic base with abductive appearance — means LLMs can assist idea generation, but outputs must be critically assessed since models cannot identify truth or verify their own explanations
- The authors address five objections and note limitations of the analysis
Related event: Sandberg and Booch Debate Whether LLMs Can Evolve Theory(2 posts)→
More from AGI Musings
- Uber burned its annual AI budget in 4 months with no visible improvement — yaroslavvb · 2026-11-23
- Post-training may exceed 75% of FLOPs, undercutting perplexity-based safety bars — stochasticchasm · 2026-11-23
- FDA greenlights first human trial of age-reversal gene therapy — Dr_Singularity · 2026-11-23
- Gary Marcus doubles down: calling out overhyped LLM capabilities is engineering rigor — GaryMarcus · 2026-11-23
- AI safety researcher slams doxxing in the community: don't burn the commons on turn one — sjgadler · 2026-11-23
- AI proofs are a different kind of brute force: fast enough to skip inventing new math — moultano · 2026-11-23