The latest AI report discusses the lag of open-source models, such as Alpaca, which have mostly imitated ChatGPT. The report proposes that the open-source community should consider improving LLM training and RLHF to make progress. However, some AI influencers have overhyped their capabilities, resulting in failed AutoGPT experiments. Despite this, the report acknowledges that more research is needed in this area, and smaller-scale tasks may help agents excel and, eventually, combine these agents. Meanwhile, a new research paper from Google DeepMind introduces a method to create advanced computer programs (LLMs) that create their own tools using a high-power model to make tools and a smaller model to use them.
https://theaireport.substack.com/p/ai-report-4-autogpt-and-open-source