New LLMs May Make Probabilistic Programming Viable Again

Researcher xuanalogue discusses using LLM logprobs as priors and likelihoods in probabilistic programming, noting that fine-tuning has degraded calibration since the GPT-3.5 era, and suggests newer models may restore this capability.

2026-09-28 ~ 2026-09-28 · 2 related posts

• xuanalogue: models like Jev make LLMs usable as priors for probabilistic programming again — xuanalogue · 2026-09-28
• Thesis chapter argues LLM logprobs can power probabilistic programming again — xuanalogue · 2026-09-28