A new advancement by Google DeepMind and the University of Southern California is set to revolutionize large language models (LLMs).

Their 'SELF-DISCOVER' framework substantially improves reasoning tasks by up to 32% over traditional methods. By empowering LLMs to self-generate reasoning structures, this approach guides models like GPT-4 and Google's PaLM 2 through complex problems more efficiently.

This innovation is a significant step towards endowing AI with human-like problem-solving abilities and could pave the way for tackling broader and more sophisticated challenges. The adoption of self-discovery methods reflects an ongoing trend in enhancing AI reasoning and general intelligence potential.