>> SIMA’s results show the potential to develop a new wave of generalist, language-driven AI agents. This is early-stage research and we look forward to further

>> SIMA’s results show the potential to develop a new wave of generalist, language-driven AI agents. This is early-stage research and we look forward to further building on SIMA across more training environments and incorporating more capable models. From the very beginning DeepMind has promised that its advances in getting agents to operate in virtual environments (mainly, games, e.g. Atari, or chess-go-shoggi) will lead to agents capable of operating in real-world environment with comparable skill and capabilities. So far, it hasn't happened and it keeps not happening. Now they added some super-trendy natural language stuff in, and it will keep not happening. If the point is to get computers and robots to not be dumb in the real world, that's all just a big waste of time and money.