Comment by pwatsonwailes

Vision language models. Basically an LLM plus a vision encoder, so the LLM can look at stuff.