Probably raised $9 million（約14億円） to build AI systems that catch errors before users see them, using a validator paired with smaller models to achieve 99.99% accuracy.

What happened: Probably, a startup founded by Peter Elias, secured $9 million（約14億円） in seed funding from Andreessen Horowitz. The company built a data science tool that pairs AI models with a deterministic validator system—if the AI's answer doesn't match the underlying dataset, the system bounces it back. The AI has been trained against this validator, and results come with citations and audit trails.

Why it matters: AI systems frequently produce hallucinations and factual errors, and the industry is still figuring out how to catch them reliably. Probably's approach lets it run on models that are 'four classes weaker than the frontier models,' which means it can run on a desktop computer instead of a data center, significantly reducing token costs at a time when those costs are rising and customers are reassessing AI budgets.

What to watch: Elias frames the insight as 'the better your harness engineering is, the weaker the model can be'—meaning the system architecture, not raw model power, drives accuracy. He sees the same engine extending beyond data science to 'precision-sensitive use cases' like accounting or medical services. Notably, he points out that large AI labs have not attempted this approach, possibly because they profit from having users correct model errors repeatedly.