A global workspace in language models
anthropic.com
A global workspace in language models
1–10 of 218 posts
Re: A global workspace in language models
#2Without using the term, they are using an information geometric approach.
Re: A global workspace in language models
#3It would be really cool if they could expose this information to customers somehow. Imagine:
- having a log of the most prominent J-space tokens during your customer support chatbot's interactions with a user, so you can have more introspection into why a particular outcome happened
- being able to detect certain thoughts associated with undesirable behavior (hallucinations, overstepping authority, lying, etc.) and trigger some sort of remediation (e.g. upgrading to a better model, redirecting to a human, forcing tool calls)Re: A global workspace in language models
#4Is it scaling up of https://openreview.net/forum?id=w7LU2s14kE with some changes on where this method is applied?
Re: A global workspace in language models
#5It would be really cool if they could expose this information to customers somehow. Imagine: - having a log of the most prominent J-space tokens during your customer support chatbot's interactions with a user, so you can have more introspection into why a particular outcome happened - being able to detect certain thoughts associated with undesirable behavior (hallucinations, overstepping authority, lying, etc.) and t…
Presumably the rationale for the decision to abridge the thinking traces will ensure that they don’t; if this is real (and there’s no good reason to trust that it is yet) then it is the secret sauce.
Re: A global workspace in language models
#6I’m confused where in the weights the jspace is.
Re: A global workspace in language models
#7Maybe model performance could increase dramatically if we found a way to scale this up.
Re: A global workspace in language models
#8As someone who is not an AI researcher, the paper itself is way over my head.
More interesting was the independent commentary paper they linked near the bottom: https://www-cdn.anthropic.com/files/4zrzovbb/website/cc4be24...
Neel Nanda (of Google Deepmind - his part begins on page 33) discusses his opinions on the paper, and the small-scale replication he performed on an open-weight model.
Re: A global workspace in language models
#9I’m confused where in the weights the jspace is.
Tokens that are activated but not present in it's output maybe?
I too have confusion.
Re: A global workspace in language models
#10Without using the term, they are using an information geometric approach.
But J-Space is much catchier. This is not a scientific paper, it's a promotional essay.