DeepMind, acquired by Google in 2014, has achieved remarkable results in software that uses reinforcement learning to master simple video games. In 2016, AlphaGo defeated Go champion Lee Sedol, showing that machines can become proficient at complex games. To train large language models like OpenAIs GPT-4, vast amounts of handpicked text is fed to machine learning software. Reinforcement learning based on human feedback is used to fine-tune the AI model’s performance. DeepMind researchers are exploring ways to enhance their large-scale language model technology with ideas from other areas of AI, such as robotics and neuroscience. However, some AI experts see the fact that language models learn about the world indirectly through text as a limitation. DeepMind CEO Demis Hassabis is tasked with accelerating Google’s AI efforts while managing potential risks.
Related Posts
