Kevin Murphy's Model Discovery Agent: LLM-assisted Bayesian experiment design hits SOTA
sirbayes · x · 2026-08-26
Kevin Murphy posted a major update (v4) to his paper "Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models" (arXiv:2608.09696), along with a talk video from U Toronto. The premise: science aims to learn mechanistic world models that answer interventional "what if" questions, and since experiments are expensive, learning must be data-efficient.
MDA combines three ingredients:
- a novel SMC³ algorithm with three nested levels of sequential Monte Carlo (models, parameters, latents);
- an LLM that proposes new models when the current hypothesis space is insufficient (M-open Bayesian inference);
- an experiment designer maximizing Value of Information.
It sets new SOTA on DPbench, CHEMbench and boxing, and introduces HHbench, a new stochastic single-neuron electrophysiology benchmark.
More from Research
- Oversubscribed London AI x Science hackathon maps researchers' biggest software bottlenecks — ersatzben · 2026-10-15
- Hannah Fry: How OpenAI has triggered an existential crisis in mathematics — AnnaCiaunica · 2026-10-15
- Watching AI evaluation shift from offline benchmarks to online eval — vboykis · 2026-10-15
- StartLux-Decision debuts RL-trained calibrated decision models, 27B scores 72.8 — startlux-models · 2026-10-15
- A third of all math preprints in 9 months were written with AI, researcher says — IgorCarron · 2026-10-15
- AAAI/ACM AIES paper interviews 15 data scientists on why responsible AI practice is hard to sustain — ChinasaTOkolo · 2026-10-15