AI That Sounds Right Versus AI You Can Prove
カートのアイテムが多すぎます
カートに追加できませんでした。
ウィッシュリストに追加できませんでした。
ほしい物リストの削除に失敗しました。
ポッドキャストのフォローに失敗しました
ポッドキャストのフォロー解除に失敗しました
AI That Sounds Right Versus AI You Can Prove
-
ナレーター：
-
著者：
Send us Fan Mail
Scott Cohen founded Jaxon.AI in Boston to build deterministic, mathematically provable AI verification for insurance, financial services, defense, and life sciences. He got there through an Air Force project where the use case had to work—and where the variability of large language models wasn't sufficient.
In this episode: why LLMs are token predictors that return the highest-probability answer rather than the correct one, why asking the same question twice gets you two answers, and why retrieval-augmented generation and LLM-as-judge are probabilities stacked on probabilities. Scott's alternative is symbolic reasoning — rule engines where a thing either is or isn't. That approach is what DSAIL does.
Also: growing up in a family of inventors behind Sweetheart Cup and the flexi-straw, what training as a chef at the Cambridge School of Culinary Arts taught him about mise en place and running a company, and the costliest hire he ever made — data scientists when he needed builders.
Guest: Scott Cohen, Founder and CEO, Jaxon.AI — jaxon.ai
Recorded 7 July 2026.
Support the show