Andy Halliday

Or what's the outcome of their having identified the location of that in a deep neural network or in the processing of a deep neural network?

Yeah.

is the import that we now have better transparency of what's going on in the innards.

And now we can control that.

So that's, I placed it as a concept in my mind that I go back and figure out.

But you know, that's what I believe it is, is that they're identifying something that, that gives them more controllability and,

over the outcome of a model as it progresses in its intelligence.

Yeah.

Okay, I'm going to go on to something else here.

Now, this is in the domain of discussions about Claude Code and particularly use of Claude co-work as well.

And on the subject of Claude's skills.

Now, skills was one of these architectural...

notions that Anthropic came forward with that really created the idea of

autonomous agents being able to go out and collect the skill that would like on the fly, improve its capabilities in a needed space.

So you can build skills yourself.

And that's basically just a file with a prompt in effect, a markdown file that describes how to do a certain kind of function or operation.

And basically,

Then, you know, we had this idea that, well, wouldn't it be great if there were, you know, an agent accessible search engine for all the skills that are published out there?

that you can go and grab as necessary or your agent can go out and grab when it doesn't have a full understanding of what needs to be done around some requests that you have.

Suddenly it finds itself in an area that you haven't published a skill, but now there should be something out there.