なぜFlintは「7」と答えるのか――主流のLLMにはできない理由
オーストラリアのスタートアップSpringboardsが、標準的なLLMの予測可能な「グループシンク」を避けるようFlintをトレーニングし、ユーザーにより創造的な出力を保証している。
springboards launched flint, a large language model specifically trained to generate a wider range of responses than mainstream models, which often suffer from predictability or 'groupthink'. while standard llms frequently return the number '7' when asked for a random number between 1 and 10, flint aims to offer more creative and diverse answers for tasks like brainstorming or travel planning. this development highlights a shift in ai focus from raw capability to control and output diversity, addressing a key limitation of current foundation models.