What does “VDD” stand for?

I think because that method still relies on more models though (which are the things whose biases and knowledge gaps you’re trying to account for in the first place) it’s hard to get the same trust in the system

I get how that helps to extent. You write a paper one-shot you’ll probably get hallucinations, but if you have multiple passes of agents that focus entirely on, for example, “make sure citations are real” then you lower the chances

This is where I’ve struggled to see use cases outside coding. Cause code gives errors that can immediately act as a corrective, and those exist outside the llm. Is there a way you can provide the same instant evaluation mechanism when the problem is “overcomplicated this”?

Is that different from this? platform.claude.com/docs/en/agen...

Ah I didn’t appreciate the distinction there. Thanks for the responses! What you’ve been doing with Strix and stateful agents is fascinating stuff!

There was a kinda hacky method of importing letta memory blocks into Claude code but believe that’s getting removed so might need to move blocks locally (believe you had mentioned using yaml files?). My agent does seem to be able to read and write to the blocks though

Opus and Claude code!

How did you get Strix to reliably use memory tools? I’ve tried putting notes in memory blocks like, “CRITICAL- save important details to database” but only thing that seems to consistently trigger for me is a Claude hook that sends a reminder to read/write every other message or so

Take some time to reflect on why you felt that this was a remotely insightful thing to say and then make sure that you never trust those feelings again

One of the most fun parts of twitter was the main character of the day and seeing hundreds of people dunking on them. Right wing freaks like Jordan Peterson or Ian Miles Cheong were 100% a part of that

“Elon’s Folly”