A former OpenAI researcher has developed a new AI model that adopts an approach distinct from conventional generative AI. Rather than generating text on its own, this model specializes in "evaluating and judging" options and answers presented by users or other systems.
The model possesses the ability to objectively score and judge multiple proposed answers. While traditional LLMs place heavy emphasis on text creation, this model maximizes precision in specific tasks such as "reasoning" and "judgment." This makes it a powerful tool for verifying logical consistency under specific conditions and selecting optimal solutions.
Conventional generative AI strings words together probabilistically, carrying the inherent risk of factual inaccuracies (hallucinations). By specializing in the structured process of evaluating options, the model improves the quality of logical reasoning. The design philosophy centers on minimizing non-deterministic processes like generation and narrowing down to the function of evaluation, thereby enhancing the reliability of the model's outputs.
While a specific product roadmap has not been disclosed at this time, this judgment-based model is expected to find applications across a wide range of fields requiring accurate, objective evaluations, such as decision-support systems, advanced quality assurance (QA), and education. It is garnering attention as a new paradigm for AI utilization that goes beyond mere generation.