“DeepMind's AI Co-Scientist project has demonstrated that language models can provide meaningful critiques of ideas proposed in natural language, serving as an alternative to strict evaluation functions.”