Startup Probably raises $9M to make AI as accurate as ordinary code

US startup Probably has closed a $9M seed round. The money will go toward AI systems that prevent hallucinations and factual errors before an answer reaches the user. The main goal is accuracy comparable to deterministic systems, that is, ordinary software code.

Why hallucinations are more than an inconvenience

Large language models regularly "hallucinate": they confidently state false facts, cite studies that do not exist and invent quotes. By various independent estimates, even the most advanced commercial models get 10–20% of specific factual claims wrong. Meanwhile, it is extremely hard for a user to tell when the model is right and when it is making things up, because it sounds equally confident in both cases. This creates a systemic risk that limits how far AI can be used in real business scenarios.

Companies are forced to choose between two bad options: hire people to verify every AI answer, or restrict models to tasks where an error is not critical. For medicine, law, financial compliance and the public sector, this effectively means autonomous AI systems are unusable, because the cost of a mistake is too high.

A new bar: deterministic accuracy

Probably is setting a nontrivial bar: to reach the accuracy of deterministic systems. That is fundamentally different from how today's LLMs work. A deterministic program always produces the same result for the same input: a calculator will never "decide" to invent an answer to 2 + 2. A language model works differently: it generates probabilistic text that may sound convincing but is not obliged to be correct.

The Probably team is building an architecture in which the AI does not just generate an answer but verifies it before sending. Key principles:

• A clear separation between reliable knowledge and the model's assumptions
• Verification of every factual claim before it reaches the user
• An explicit "I don't know" instead of an invented fact
• Transparency: the system explains the source of each claim
• Zero tolerance for factual errors in the final answer

This approach is the opposite of the direction most LLM development has taken in recent years. Models were fine-tuned with an emphasis on "helpfulness" and "persuasiveness", which often conflicted with accuracy.

Market and competition

Reliable AI is currently attracting active venture capital. Players in this niche include Vectara, Cohere with its focus on enterprise RAG architectures, a number of stealth startups and academic projects at major universities. The big players, OpenAI, Anthropic and Google, are also investing in reducing hallucinations, but for them accuracy is only one of many product characteristics, not the main differentiator. Probably builds everything around a single idea.

The $9M seed round gives the team a year and a half to two years to prove the architecture works and raise its next round of funding. The company's name is a deliberate irony about the probabilistic nature of language models, and at the same time a declaration of intent to overcome it.

What this means

If Probably manages to convincingly solve the hallucination problem, it will open the way for AI into regulated industries, such as healthcare, legal services and financial compliance, where autonomous systems are unacceptable today because of their inaccuracy. For businesses, this means the chance to remove the costly layer of human verification and genuinely delegate high-stakes tasks to a machine, backed not by words but by measurable accuracy guarantees.