Maybe raise $9 million to build a more reliable type of artificial intelligence

As LLM holders become more powerful, it has proven very difficult to avoid hallucinations. Bugs appear in even the smartest models, and although there are ways to detect these bugs, the industry is still figuring out the best way to do it.

probablywhich just raised $9 million in seed funding from Andreessen Horowitz, is trying to build a more rigorous way to catch those errors.

As founder Peter Elias (pictured above) says, the company’s goal is to prevent hallucinations and simple factual errors from reaching the user, achieving the kind of 99.99% accuracy that is common in deterministic systems but much more difficult to reach with artificial intelligence. As it turns out, getting LLM students to this level of rigor requires rethinking several basic assumptions of AI engineering.

Perhaps the first product is a data science tool, designed to produce quick answers from complex data sets. Each result comes with a quote and an audit trail of how it was developed, an increasingly common practice among AI tools.

But preventing errors from creeping into those summaries requires a sophisticated system that Elias describes as a “data science mechanical suit.” LLM’s first-pass answers are checked against a deterministic validation system, which bounces any results that do not match the data set. Most importantly, LLM has been trained as an auditor, and the entire system is optimized for quick and accurate answers, the company said.

“What we learned from building this is that the better the geometry of the belt, the weaker the model,” says Elias. “If you can optimize the context enough, the model doesn’t have to work as hard to do the right thing. It’s basically an exercise in ambiguity reduction.”

This allows Maybe’s data science tool to work on much smaller AI models. Elias says the current version runs on a model that is “four classes weaker than the frontier models,” which means it can run on local hardware (i.e. a desktop computer rather than a data center), reducing a huge amount of token costs associated with using AI.

It’s a welcome idea at a time when token costs are rising and the number of clients is rising Reevaluate their AI budgets. And Elias’ idea doesn’t end with data science, as the same engine can be extended to use cases like accounting or medical services — “any accuracy-sensitive use case,” Elias says.

“I think it’s really interesting that the big AI labs haven’t even tried to do this,” Elias says. “They are incentivized not to do it, because they make money the more times you have to correct the model.”