LLMs fail basic probability coherence: P(rain)=0.7 but P(no rain)=0.2
Moh1tAgarwal · x · 2026-09-05
Researcher Suproteem Sarkar illustrates a probability-coherence failure in LLMs: ask a model the chance of rain tomorrow, then the chance of no rain — the two must sum to 1, yet the model may output P(rain)=0.7 and P(no rain)=0.2. The thread targets calibration and consistency of LLM probability outputs.
More from Research
- Anima Anandkumar's first podcast on Accelerated Understanding: 5T context and 4D physical AI — AnimaAnandkumar · 2026-09-05
- Uno pairs AR weights with diffusion LoRA adapters, beating EAGLE-3 and all diffusion LLMs — HongyiWang10 · 2026-09-05
- Self-explanation training generalizes beyond narrow hint formats to held-out evals — a_karvonen · 2026-09-05
- Two training targets from behavior investigations: counterfactual predictions and open-ended self-explanations — a_karvonen · 2026-09-05
- Anthropic Fellows train models to explain their own wild behaviors with generalization to held-out evals — a_karvonen · 2026-09-05
- Deep Learning Weekly #471: Claude Fable 5.1 launch, production-parity LLM evals, alignment paper — dl_weekly · 2026-09-05