The point: 2Brains architecturally separates language generation from fact retrieval to prevent hallucinations, simultaneously reducing energy consumption and costs.
IT journalist Robert Cringely founded the startup 2Brains in 2022 to solve the hallucination problem in AI systems through architectural separation rather than mere model scaling. The solution runs on conventional processors and breaks with the strategy of tech giants to scale computational power.
The 73-year-old journalist Robert Cringely, who began his career in 1978 at Stanford Artificial Intelligence Lab, has been working with two partners on 2Brains Inc since 2022. Cringely explained his two-year publishing break with health consequences – a heart attack and a stroke in July 2025 – as well as the founding and development work on the startup. Patents have already been filed, and the system architecture documented. The small team continues its work with Cringely’s involvement.
Cringely sharply criticizes the current scaling model of leading tech companies. They invest billions in expanding computational power and models, but follow a flawed assumption: that the hallucination problem – the generation of false information by AI – solves itself through pure scaling. As long as systems invent facts, they are unreliable for deployment in hospitals, banks, or courts. Directly addressing the hallucination problem is therefore the key to productive enterprise AI.
2Brains pursues a fundamentally different path. The architecture strictly separates the language generation component from the fact retrieval and verification component. Both processes are reconciled before the system outputs an answer. Instead of statistically guessing facts, the system looks them up systematically – making it impossible to fabricate information. The technology runs on conventional processors and significantly reduces energy consumption compared to large AI models.
Cringely succinctly summarizes the economic advantage: “The reason why 2Brains does not lie, and the reason why it is cheap, are the same reason. It looks up the fact instead of guessing it – so it cannot invent anything, and looking it up runs on a processor that sips power instead of on a chip that guzzles it.” Trust and cost-effectiveness are not contradictions, but result from a single design decision.
Source: www.it-daily.net · Published June 27, 2026
Lumi AI News — AI-assisted curation pursuant to Article 50 EU AI Act. Paraphrase and classification via Lumi News Pipeline v1.7.1.