In the rapidly evolving world of artificial intelligence, the persistent problem of AI “hallucinations”—where models generate plausible but incorrect information—remains a formidable challenge. Probably, a startup focused on enhancing AI reliability, has just secured $9 million in seed funding from Andreessen Horowitz to tackle this issue head-on. Their mission: to build AI systems with near-perfect accuracy that can be trusted for critical, precision-sensitive tasks.
Addressing AI Hallucinations with a New Engineering Approach
Large language models (LLMs) have revolutionized how machines understand and generate human-like text, but their outputs are often marred by factual inaccuracies. These hallucinations can undermine trust and limit AI adoption in high-stakes environments like healthcare, finance, and legal services. Probably’s founder, Peter Elias, argues that the key to overcoming this lies not just in bigger or more complex models but in fundamentally rethinking how AI is engineered.
Probably’s approach involves a hybrid system that combines the generative power of LLMs with deterministic validation layers. Their flagship product is a data science tool designed to extract fast, accurate answers from complex datasets. Every AI-generated response is cross-checked against a deterministic validator that ensures the output aligns with the underlying data. This “data science mech suit,” as Elias calls it, acts as a rigorous filter to prevent errors from slipping through.
Reducing Ambiguity to Increase AI Accuracy
One of the most insightful aspects of Probably’s technology is its focus on reducing ambiguity in AI prompts and contexts. Elias explains that by refining the input context carefully, the AI model itself does not need to be as powerful to produce correct answers. This means the system can rely on smaller, less resource-intensive models without sacrificing accuracy.
This is a significant breakthrough because it challenges the prevailing notion that bigger and more complex models are always better. Instead, Probably’s system optimizes the interaction between model and validator, allowing the AI to run efficiently on local hardware such as desktop computers rather than costly cloud data centers. This dramatically cuts token consumption and operational expenses, a growing concern as AI usage scales and costs rise.
Implications for AI Adoption in Precision-Critical Fields
Probably’s technology is not limited to data science applications. Elias envisions its extension into any domain where precision is paramount, including accounting, medical diagnostics, and regulatory compliance. The ability to guarantee 99.99% accuracy—common in deterministic software but rare in AI—could make AI a trusted partner in fields where errors have serious consequences.
This development comes at a time when many organizations are wary of integrating AI due to reliability concerns. By offering a system that inherently prevents hallucinations, Probably could accelerate AI adoption in sectors that have so far been cautious.
Challenging the Status Quo in AI Development
Interestingly, Elias points out that major AI labs have yet to prioritize this level of accuracy. He suggests a misalignment of incentives: companies benefit financially when users repeatedly correct AI mistakes, which drives more usage and revenue. Probably’s model flips this dynamic by striving to eliminate errors upfront, potentially reducing the need for costly human oversight.
This investor-backed gamble on reliability over sheer scale highlights a maturing AI ecosystem where trustworthiness is becoming as valuable as raw capability. The $9 million funding round led by Andreessen Horowitz underscores confidence in Probably’s vision and the growing market demand for dependable AI solutions.
Looking Ahead: The Future of Trustworthy AI
As AI continues to embed itself in everyday life and business operations, the pressure to deliver error-free outputs will only intensify. Probably’s pioneering work suggests that achieving this goal requires more than incremental improvements—it demands a reimagining of AI architecture that balances generative creativity with rigorous validation.
If successful, Probably could set a new industry standard for AI accuracy, opening doors to applications previously deemed too risky for automation. In doing so, it may help shift the narrative around AI from one of uncertainty and caution to one of confidence and reliability.