I haven't seen a good write-up on this, however it appears that large Transformer architectures are directly learning a memory representation in their parameter

I haven't seen a good write-up on this, however it appears that large Transformer architectures are directly learning a memory representation in their parameters. ChatGPT can easily return facts, and approximate summaries of facts. Unfortunately ChatGPT is unable to differentiate what it knows, vs. what it thinks it knows. I'd love to see a deep dive into how these models are representing such knowledge.