Tool usage with LLMs has been discussed for more than 1.5 years. However, it failed to make any noticeable progress until OpenAI’s Deep Research achieved a breakthrough. It is essentially a language model that can use tools, trained using reinforcement learning with the same paradigm as o1. This makes me imagine that AGI will soon be realized. Tool Usage + Reinforcement Learning will soon surpass any human individual in most non-physical tasks. If I were Ilya, I would also be astonished. It’s a social crisis.