While it would be a hilarious failure mode to encounter, this is actually a good thing!

These models already have the skills that humans were using them for, so either by training the models to use subagents or simply inlining the work done by the AI, you have a much easier time training the model to perform tasks from a human-distribution. The humans have done the work of making the human-distribution look more like an AI distribution.

Doesn't this assume that what humans are current doing with LLM agents is working out? Isn't it a bit early to bet on that to this degree?

Not when all of the marketing of LLMs is touting their abilities to do the exact thing and that is what investors are being presented.

If it is as you say, then eventually the house of cards will crumble. Then we can finally go back to work and quit being inundated with needing to use AI for everything.

Turns out if we say any word with the same inflection as sit, our dog sits!

I think it's partly because "sit" is one of the first commands they learn so if they're not sure what to do, they'll default to sit as that often gets the treat.

That's also why you teach "sit" first before, "bite the face of the person in front of me" (talking German Shepherds again)

Open access typically means authors pay a publication fee, which leads to the same result of the government paying twice and the journal profiting twice.

Almost as if they shouldn't be getting special treatment in the first place!

  insertDrakeMeme()
  nah: Actor-Critic Models
  ayy: Actor Model

In seriousness, huge fan! multi-agent systems are inherently distributed systems, and we've solved the problem of organizing distributed systems before through actors.

...Just wait until we get multi-agent systems with enough agents to need distributed consensus!

This I didn’t know!

recover()'s semantics make it so that "pointless" use like this can be inlined in a way that changes its semantics, but "correct" use remains unchanged.

Yes, maybe some code uses recover() to check if its being called as a panic handler, and perhaps `go fix` should add a check for this ("error: function to be inlined calls recover()"), but this isn't a particularly common footgun.

> ... and perhaps `go fix` should add a check for this (

This is an impossible task. For a library function, you can't know whether or not the function is defer called.

Maybe this is not an important problem. But it would be better if the blog article mentions this.

by future do you mean Future<T> or metaphorical future? :)

I see what you did there.

hey hey hey the world would be a better place if we all used JAX instead :)

There’s a known fix for SQL injection and no such known fix for prompt injection

There is one pretty simple change developers can make to protect against "prompt injection" though.