It's a memory retrieval and synthesis model, akin to human long term memory. I think it's a bit disingenuous to call true multimodal models for "language models

It's a memory retrieval and synthesis model, akin to human long term memory. I think it's a bit disingenuous to call true multimodal models for "language models". A prompt triggers a bunch of memories which gets recombined to satisfy the input. For cognition you need an agent that uses these recombined memories, a temporary scratch pad (short term memory) and an algorithm for solving the problem.