Probably raises $9M to build a more reliable kind of AI

The push to develop AI systems that reliably avoid hallucinations isn’t just about technical refinement—it’s a fundamental challenge to the credibility of artificial intelligence itself. Probably’s latest funding round underscores how high the stakes have become. Generative AI has unlocked unprecedented capabilities in language, creativity, and problem-solving, but its tendency to fabricate facts or distort reality has limited its adoption in critical domains like healthcare, law, and finance. For enterprises and consumers alike, trust hinges on accuracy, not just fluency. The company’s ambition to match deterministic systems in reliability suggests a pivot from chasing raw performance to prioritizing stability—a shift that could redefine where AI is deemed safe and useful. This isn’t happening in a vacuum. The reliability gap has already sparked parallel efforts across the industry. Tech giants are embedding retrieval-augmented generation (RAG) to ground AI responses in vetted data, while smaller startups are experimenting with hybrid architectures that blend probabilistic models with rule-based logic. Regulators, too, are taking notice, with proposals for mandatory transparency in AI-generated content gaining traction in the EU and U.S. Probably’s approach—securing $9 million to refine its methods—implies that the market sees a viable path forward where AI doesn’t just sound convincing but can be consistently accurate. Yet the path is fraught with technical hurdles. Training models to suppress falsehoods without sacrificing creativity, or to cite sources dynamically, requires breakthroughs that remain elusive. The funding signals investor confidence, but the real test will be whether these systems can scale without introducing new, subtle biases or vulnerabilities. What comes next hinges on whether Probably can demonstrate that its model doesn’t just reduce errors but does so in a way that feels seamless to users. Open questions linger: Will enterprises adopt these systems if they’re slower or more constrained? Can the technology keep pace with the evolving nature of misinformation, where even subtle distortions can have outsized real-world consequences? And perhaps most critically, will the public ever fully trust AI if it remains fallible in any capacity? As the AI landscape matures, the race for reliability may prove as consequential as the race for raw capability.