XNews

Google DeepMind Dream-RSI cuts agent calls by up to 162x

Google DeepMind Dream-RSI cuts agent calls by up to 162x

Google and DeepMind showcased Dream-RSI, a recursive self-improvement loop that replays past discovery processes to test exploration strategies cheaply. It maintained or improved quality in algorithms, math optimization, and GPU kernel engineering, reducing agent calls up to 162 times.

Source posts