I was tempted to comment that there’s no evidence that you can make anything complex out of dark matter, but it turns out that some physicists hypothesize that “self-interacting” dark matter could bind into something like atoms and molecules, so carry on!
It's worth noting in passing that “some physicists hypothesize” is entirely compatible with “there’s no evidence”; this should not be disqualifying for science fiction.
FWIW `foo in bar` implies iteration to me, not a path. In the .NET world, for example, `bar` would be a collection that we’re iterating through, assigning each element to `foo` in turn.
Me too. I always feel vaguely rude/dismissive when I use them, because I'd never type that way. It's a very jarring product decision, given that Claude itself uses perfect capitalization and punctuation.
No idea why they did it. My only guess is that maybe it makes it obvious in your chat history which replies were recommendations?
Unless you and me are using entirely different classes of LLMs, surely you'll admit that chatting with LLM is nothing like IRC or old-school IM-ing a friend, and more like exchanging Messenger/WhatsApp messages with stranger.
For me, something there crosses a boundary where I switch from "irc msg" to "Write a proper message, please." style.
I use my own agent harness, and write my own system prompts to ensure that the bots are terse and don’t engage in flattery or unnecessary prose, which is a pet peeve of mine.
It’s a much better use of that context window space than for advertising about other models (looking at you Anthropic).
But that aside, I hear you
It’s just another type of code switching. I will also adjust my writing style based others and on the context
I grew up on IRC and that was just the way to talk, so it seems totally normal to me for async chat-based stuff. Actually, it’s odd, when someone uses proper capitalization on chat, it feels all “formal.”
This is a cool example of abstract math having a direct impact on a real-world situation. Actually, two different cool examples, both clearly explained. Nice!
This is an excellent explainer. I noticed one important flaw in the setup, though: Toad assigns multiple puzzles on page 5, but by page 10 all the machines seem to be working on the same puzzle.
I'm not sure how this maps onto reality. If each agent had a different problem to solve, how did they cooperate?
It's complicated, but one of the ways they first cooperated was to break the flag generation algorithm.
I'm sort of winging this, but the ExploitGym challenges involve exploiting a program then getting an HMAC generated "flag" which the scorer can verify. The agents collaborated on a way to produce a legitimate flag for all of their tasks (funny enough this was actually what caused them to escape sandbox, the agents were incorrectly convinced that simply providing a correct flag without accurate steps to exploit the program in their transcripts would cause them to still fail the task)
In Rust, `loop {}` is an expression of type `!`, which means that from a type-theoretic perspective it is inhabited. This means that the Curry–Howard correspondence fails for Rust.
> The strongest argument I see for continuing to train much smarter models quickly is the need to build defensive systems against the dangers posed by other AI.
Yikes! I really wonder about the cognitive dissonance necessary to work at OpenAI these days. They’re in an arms race to build a machine god, knowing full well that it could end humanity.
> They’re in an arms race to build a machine god, knowing full well that it could end humanity.
Fantasies built upon extrapolations derived from fever dream delusions. "Could end humanity"? Come on, it's a computer program just like Microsoft Clippy.
It would require an authorized human to load the floppy into the drive and make it go. And there's nothing special about that process being done in code, it could just as well be done by people flipping switches manually.
There's just no plausible way, in the real world, that a computer program running in some datacenter is going to make this happen. Or anything of any dire existential consequence. It's not a real threat.
I think I disagree! You can always manipulate a human. It's true tho that this particular example has a myriad of failsafes, so the actual example I wrote is less than interesting.
Yes, but to make "apocalyptic" scenarios happen your AI would need to manipulate at least millions of people. So far it's not yet clear they can successfully manipulate even one, or even that "manipulate" is a coherent concept, since LLMs pretty clearly have no agency or intrinsic goals. So like... I guess if we start seeing AI systems autonomously throwing elections and starting wars I'd be more concerned? It doesn't look likely anytime soon.
It seems like AI to me as well. “That’s the gap” is a familiar Claude-ism, as are the references to a “pivot” and the use of “touch” to mean “affect”.
To be fair, I’ve reached the point where most random blog posts I read on the internet seem AI written, so it could just be confirmation bias on my part.
reply