- Google is trying to make Gemini more efficient - Google researchers are looking for new ways to make progress - The slowdown in the development of its AI was previously reported by OpenAI - Google has not yet achieved performance improvements - More power and training data have not been able to help - That said, for past versions of Gemini, this approach worked - Google's Gemini models now lag behind OpenAI's GPT - It's about the number of developers and users - The resource advantage didn't help Google beat OpenAI - OpenAI was able to invent new methods for AI training - In particular, OpenAI uses reasoning models, etc. - Google has now decided to follow its competitor's lead - It has also focused on manual improvement of models - i.e. changing the “hyperparameters” of its models