Posts on what we actually think.
Essays on conformal prediction, social inflation, and where generative AI helps and where it doesn't. Written by the team.
Showing 14 posts · David H. Silver
Clear filtersWhy Comparable Verdicts Overstate Settlement Exposure
Public trial verdicts are terrible predictors of settlement value. The data is systematically biased toward extreme disagreement, leading models and adjusters alike to radically over-reserve.
When AI Should Say It Doesn't Know
Language models are built to generate plausible text, not calculate risk. When an AI offers a single dollar figure for a complex claim without quantifying its own doubt, it is not predicting. It is guessing.
Training on Known Outcomes
Large language models are built to talk, not to calculate risk. Relying on them to predict claims outcomes conflates reading comprehension with mathematical forecasting.
Benchmarking Litigation Outcome Prediction
A point prediction for a complex liability claim is mathematically meaningless. True litigation forecasting requires separating the extraction of text from the calculation of risk, delivering calibrated ranges rather than brittle guesses.
Want to talk to an executive?
Press, partners, investors, candidates — the inbox is monitored. Tell us who you are and we'll route it to the right person within two business days.