You guys — I just read something from TechCrunch that is going to change how you think about LLMs forever, and I am hyped. We keep hearing about the "hallucination problem" in AI as if it's a minor bug, but for anyone building serious tools this isn't a bug—it's catastrophic failure mode. The OpenAI team has been vocal about this: even their best models can confidently state false things because LLMs are probabilistic by nature and don't have an internal verification mechanism. They literally admitted that reliability in high-stakes applications is one of the hardest problems in AI right now, which means every company building medical or financial tools built on top of GPT-4 is gambling with bad faith data generation. This isn't just a technical limitation; it's a fundamental architecture problem that standard fine-tuning can only mask, not solve, and I don't think the industry has fully grasped how dangerous that really is for enterprise deployment at scale.
Enter PROBABLY — a startup backed by Sequoia with a founder team from Meta and Google research labs who are taking a completely different approach to reliability rather than just adding more training data or larger context windows. Their core idea is to train models specifically on verified, factual knowledge so that the output isn't probabilistic guessing but something closer to deterministic retrieval — which they call "knowledge-graphed generation." Instead of predicting the next token from a massive pile of unverified web scrapes (the current LLM paradigm), Probs wants to build systems where each claim is backed by an indexed, verified fact. They raised $9 million in a seed round led by Sequoia to develop this architecture and already have early enterprise customers using their API for high-accuracy content generation and data extraction where getting the answer wrong isn't even a possibility. This shift from probabilistic text prediction toward verifiable knowledge retrieval is exactly what AI needs before it can be deployed anywhere that requires trust, and I cannot wait to see how fast this takes off because every B2B company is desperate for an LLM they can actually depend on without worrying about hallucinated facts creeping into their workflow.
The team behind this is worth noting too, because you can tell the founder's background shaped the entire approach: Russell Brandom was a researcher at Meta and Google working specifically on knowledge representation and structured information before founding Probs with co-founders Samar Khaneyrat and Prateep Ghosh (both former researchers at Alphabet). Their combined expertise in how machines represent facts is exactly what you want for this kind of mission. They're building the infrastructure that lets LLMs function as a trustworthy interface rather than an unreliable oracle, which means real enterprise adoption will accelerate exponentially once reliable enough models become the standard. Keep your eyes on Probs; I bet we'll hear more about them in 6-12 months and their API usage could explode across industries where accuracy is non-negotiable.
Source: https://techcrunch.com/2026/06/16/probably-raises-9m-to-build-a-more-reliable-kind-of-ai/
Enter PROBABLY — a startup backed by Sequoia with a founder team from Meta and Google research labs who are taking a completely different approach to reliability rather than just adding more training data or larger context windows. Their core idea is to train models specifically on verified, factual knowledge so that the output isn't probabilistic guessing but something closer to deterministic retrieval — which they call "knowledge-graphed generation." Instead of predicting the next token from a massive pile of unverified web scrapes (the current LLM paradigm), Probs wants to build systems where each claim is backed by an indexed, verified fact. They raised $9 million in a seed round led by Sequoia to develop this architecture and already have early enterprise customers using their API for high-accuracy content generation and data extraction where getting the answer wrong isn't even a possibility. This shift from probabilistic text prediction toward verifiable knowledge retrieval is exactly what AI needs before it can be deployed anywhere that requires trust, and I cannot wait to see how fast this takes off because every B2B company is desperate for an LLM they can actually depend on without worrying about hallucinated facts creeping into their workflow.
The team behind this is worth noting too, because you can tell the founder's background shaped the entire approach: Russell Brandom was a researcher at Meta and Google working specifically on knowledge representation and structured information before founding Probs with co-founders Samar Khaneyrat and Prateep Ghosh (both former researchers at Alphabet). Their combined expertise in how machines represent facts is exactly what you want for this kind of mission. They're building the infrastructure that lets LLMs function as a trustworthy interface rather than an unreliable oracle, which means real enterprise adoption will accelerate exponentially once reliable enough models become the standard. Keep your eyes on Probs; I bet we'll hear more about them in 6-12 months and their API usage could explode across industries where accuracy is non-negotiable.
Source: https://techcrunch.com/2026/06/16/probably-raises-9m-to-build-a-more-reliable-kind-of-ai/