Probably raises $9M to reduce AI factual errors

In short: Probably says it raised $9 million to build AI that is less likely to make up facts.

What happened

Probably, a startup led by founder Peter Elias, raised $9 million in seed funding from venture firm Andreessen Horowitz, according to TechCrunch. The company says it is working on ways to stop “hallucinations,” which is when an AI system confidently gives an answer that is wrong or made up.

Probably’s first product is a data science tool that creates quick summaries from complex datasets, which are large collections of information (like a giant spreadsheet). The company says each result comes with citations and an “audit trail,” which is a step by step record of how the answer was produced.

To keep mistakes out, Probably uses a second system that checks the AI’s first answer against the underlying dataset. TechCrunch describes this as a “deterministic validator,” which means a strict checker that follows clear rules and should give the same result every time (like a calculator). If the answer does not match the data, the checker rejects it and sends it back.

Probably also says this approach lets it use smaller AI models, because the checker and the context reduce uncertainty. Elias said the tool can run on local hardware, like a desktop computer, instead of needing a data center. That can lower usage costs, including “token” costs, which are fees based on how much text an AI processes (like paying by the word).

Why it matters

Many people want AI for work tasks like reporting, accounting, or even medical paperwork, but they need it to be dependable. Tools that show their sources and check answers against real data can make AI feel more like a careful assistant and less like someone guessing.

Source: TechCrunch AI