ESRogs comments on Alignment As A Bottleneck To Usefulness Of GPT-3

[ESRogs]

Have we ever figured out a way to interface with what something has learned that doesn’t involve language prompts?

You might be interested in some of Chris Olah’s work on interpretability. For example, this.

EDIT: Or even just the example of sampling from the latent space of a variational autoencoder should count, I would think.

[Past Account]

[Deleted]