Probably raised $9M to curb 'hallucinations' and boost AI accuracy

Friends, I'd like to share an AI update: startup Probably raised $9M for a system that prevents hallucinations and factual errors in LLMs.

- Offers a data-science tool: answers with sources and an audit trail.
- First check uses a deterministic validator; mismatches are discarded.
- The model is trained alongside the validator and can be much weaker than frontier models, enabling local deployment.
- Applicable to accuracy-sensitive domains: accounting, healthcare, etc.

Why it matters: reduces production errors and token costs.

Should large labs scale this "harness" approach?