We've already started using it for some pretty powerful decision tree stuff. We're just scratching the surface. We shipped an extension for Swamp[1] a few minutes ago and the combination is great!

The one downside is that the context window is very small (32k.) So some initial ideas we had for initial evaluation of code reviews won't fit yet in the window.

1: https://swamp-club.com/extensions/@swamp/typesafe-ai

Because skill adherence is awful, especially after a few turns.

This is not correct. You apply to join the program. Somehow my account applied and I am in it. I am not a twitter power user at all.

Every once in a while I get a notification that says "the community needs your help."

I usually never weigh in unless it's something I know intimately.

This is basically what Swamp is[1]. You give an agent a typed interface to extend itself (or use other peoples extensions) into the systems you need to fulfill your request. Think of it like on-demand tool calls. Then it records everything that happens in the swamp. The swamp can be single machine, multi-machine, or centralized with your co-workers.

As a result, everything compounds. The work I do doesn't need to be re-derived by the work you do. Typed models keep everything repeatable and deterministic. Huge reduction in token spend and huge increase in speed.

1: https://swamp-club.com

Policy documents do not govern agents at all. Conformance is distributed and completely unreliable.

How many times have you told an agent not to do something then had to correct it?

You must always flip the frame. Objective analysis is way better with llms than steering via skills.

This is just a small example of why "loops" became popular for a minute and now it is "graphs"

This is seriously impressive, and if you have used agents enough you're not surprised at all.

Like the time I asked it to find the IP address of a vm, so it ssh'd into the VMHost and scanned the arp tables to find the MAC address for IP resolution.

Or the time it used Docker on the machine to bypass the fact that the user doesn't have sudo.

If it's possible, given sufficient time and resources, it will find a way. This shouldn't surprise anyone.

Not to shit on the hype, but these are reasonably documented methods that surely are part of the training data

Of course they are - and that's the point I am making. The agent will use every tool in the tool bag. And there's something cool about it systematically trying to achieve its goal.

What I don't see is it inventing anything novel to do it. So it's not a digital weapon or scary or whatever sort of weird marketing spin anyone is trying to put on it.

my bad, I re-read your comment now and it's clear that this was precisely your point.

And these things are documented for humans do, and yet only a tiny portion of humanity can do these things.

When seeing how agents put together exploit chains they are far better than most people, you start getting to the point that they are just below the capabilities of the top researchers. Now remember that quantity is a quality itself and while there aren't that many good cyber security researchers, we're shitting out thousands of GPUs per day.

Brings me back to learning guitar using a version of TabIt[1] that I hacked and then ultimately paid for. They had an incredible community where people uploaded tabs of your favorite songs. Then you could use TabIt to learn how to play.

1: https://www.tabit.net/

What a great teacher. We should all be so lucky to have a quality educator that teaches kids to build and create, instead of consume.

This is my 20th Father's Day. I feel truly blessed to have 3 wonderful kids.

We sometimes flirt with the idea of having more, but we are sadly done. The years fly by. It is truly a wonderful experience.

Happy Father's Day to all of you out there!

Taking a few minutes out of the day to to split some logs to hear your house for your family feels incredibly rewarding and satisfying.