Can AI automate AI R&D yet?

AI developers are testing whether current AI models can independently discover novel machine learning techniques comparable to human-developed innovations. Early results from InnovationEval show that frontier AI models make little progress on end-to-end research tasks despite using thousands of dollars worth of GPU time, suggesting that automating AI R&D remains a significant challenge. The evaluation measures whether AI can devise ML innovations that match the performance improvements achieved by human researchers, providing a realistic measure of AI's capability to conduct independent research.