Probably raises $9M from a16z to keep LLM hallucinations from reaching users

Probably emerged with a $9M seed from Andreessen Horowitz to push LLM accuracy toward the 99.99% threshold typical of deterministic systems. Founder Peter Elias describes the approach as a "data science mech suit": every LLM answer runs through a deterministic validator, and the LLM has been trained against that validator to reduce ambiguity in context. The first product is a data science tool that returns answers with citations and audit trails over complex datasets. Because the harness handles validation, the underlying model can be "four classes weaker than frontier models," letting it run on local hardware and slashing inference cost.

View full digest for June 17, 2026