Sakana AI has created a language model that can improve its own performance by using a novel technique called the MASS method. This approach enables the model to solve tasks, evaluate its results, and refine its workflows autonomously.
According to The Decoder, after two rounds of this iterative process, the self-improving model achieved a nearly 70 percent win rate against its original base model. However, the method also carries the risk of reinforcing the model’s blind spots as it iterates on its own outputs.
This development is particularly relevant to Japanese markets, where advances in AI-driven language processing could enhance trading algorithms and market analysis tools across FX, crypto, and equities sectors.