An Anthropic researcher has revealed insights into self-improving AI systems that have enhanced performance on ten benchmarks without negatively impacting overall performance. This development is part of Anthropic’s broader initiative to document internal capability gains and measurement limits. The unreleased « Model 2 » reportedly outperforms the current Mythos 5 model. These advancements suggest Anthropic’s increasing competitiveness in the AI sector.
Source: Read the original article
