qarl

> I’m not saying AI is not capable.

Well... except you did:

> without providing any benefits.

> qarl: You do realize the purpose of the parody Peter Molydeux account is to make light hearted fun of Peter Molyneux, right?

HEH. No, I did not understand that at the time - I appreciate your explanation.

HA! It just solved a ton a problems for us - but that makes you mad?

But yes, I agree, there needs to be a solution for the fact that we don't need to work anymore. And it needs to be more like The Jetsons than 1984.

We'll probably have to revolt to get the changes we need. I don't see Elon giving up his power without a fight.

> can reduce the risks inherent to LLMs, to a really remarkable degree

To an arbitrary degree.

Just like all of science. Reduce the error to the desired margin.

> but simply dumping proofs on the math community

So you'd prefer if they kept their work secret? Or you don't want them working on these problems at all? Or they should be required to do the work the way you want them to?

I'm not clear what you see as a better option than dumping.

Heh.

Math? Everyone is talking about math, with math-centric discussions and solutions. I think the obsession with math is simply to distract ourselves from the fact that it's coming for every occupation.

None of these essays even remotely consider it. Denial is a helluva drug.

Hey Brad - I recently spent some effort creating a library of animating text effects and slide transitions. Similar idea, but hardware.

Anyway, feel free to rip anything you like:

https://openmarquee.com/demo/

https://github.com/qarl/openmarquee

> LLMs have been forced into our lives

Anyone who hasn't had his head in the sand has seen this coming for the last 60 years. Why do you think Turing created his test?

At this point the stubborn refusal to see it has become laughable.

> The "apparent" hypocrisy dissappears the moment you think about it at all though, as I explained above.

Yeah... I think we'll just need to agree to disagree, friend.

I just find the apparent hypocrisy amusing.

I hear an endless stream of anti-AI nonsense daily - and this appears to be yet another example of it.

> You didn't build a thing, dude. You had a machine build it for you.

But when agents hack another company, it's not the LLM that did it but the guy who asked it to.

Strange contradiction.

Well - what I described is also called a clean-room. But sure, I imagine it wouldn't be hard to do with screenshots.

I truly wonder what's to become of copyright.

I gave my agent your exact prompt and he came up with a pretty cool idea, and I pasted it here.

So of course HN deleted it.

(It was a language game about forgetting a language instead of learning a language.)

Heh - if not - you just gave them the idea.

I dunno - I just asked my agent what he'd make:

  You start fluent in your grandmother's language and, over 25 years of play,
  lose it word by word.
  - New: language games are about learning. This one runs that process backward,
    so forgetting is the main mechanic. You never fail a conversation. You just
    have fewer words to say it with, and Nana switches to slow English to meet
    you.
  - Closest games: Venba, where the loss is scripted, and Chants of Sennaar,
    where you decode languages instead of losing one.
  - Moment of play: you're 31 in an airport, on the phone with Nana, with seven
    word tiles left. She says "Okay. Good. You eat?" and you watch a tile
    dissolve.

Thank you. The tsunami of downvotes seems to have stopped.

> i don't think llm's are capable of designing new games

I invite you to examine my clean-room implementation of Animal Crossing, coming out next week. Infinite expanse and multiplayer.

I'm not sure what you mean by "new games" - but substantial changes to existing games are extremely easy.

Also - the animals will be LLMs - so when you talk shit about them they'll remember and get revenge.

If he used a clean-room technique then yes it was.

> perhaps that decompiled source was already in the training set

Nope, it works for all games. Unless you think all game code has been secretly stolen.

Between the people yelling "you didn't do this" and the people yelling "stop trying to shirk responsibility saying the LLM is doing this" a guy can't catch a break.

Given the way he implemented it - I suspect it is optimized exactly like the original.

It sounds like your agents were allowed to directly read the old code and implement it as new code.

Technically this is a violation of the rules for reverse engineering. You need a gap between the agents that look at the old code and the agents that implement the new code. With an English spec connecting them.

Interestingly, that's also easy to do with agents. And apparently legal. Our entire system of software copyrights is hanging in the balance.

Strange days.

EDIT: I heartily apologize for citing the rules for a clean-room implementation. I shall never try to explain anything to HN again.

To be fair - there are statements in math that are independent of the axioms. For those statements, the universe you find yourself in can pick either version (true OR false) and still be consistent.

See also: noneuclidian geometry and axiom of choice.

> is that enough to flag these posts?

No it's not. Especially because this phenomenon is going to become commonplace. It is very easy to spin up two agents - one to look at the original code and write a spec for it - and the second to implement that spec. That's what's required for "legal" clean-room reverse engineering.

At the very best - these flags are sticking one's head into the sand.

> When they say ‘clean room’. What do they mean by that exactly?

I don't know what they mean - but I can tell you what I mean when I do the same thing.

The new code is written without looking at the original from an English description of the software's behavior. According to the rules of reverse engineering, this is allowed.

Of course, the problem is that agents can now do this trivially. So.

Why is this story being censored?

A clean-room reverse engineered Photoshop is huge and important news.

Are we hiding from reality now?

The thing is - agents are fantastic for creating those behavioral observations, too. Describe in English is not a problem for a language model.

I've got 75% of Animal Crossing done the same way. I assume Nintendo will try to kill me... but I am following the rules of reverse engineering. Because it's super easy to do now.

The world is a runaway train. Who knows how it will all end.

Aw man... now my Animal Crossing clean room implementation won't be a huge surprise.

Dang it. Gotta get one week faster.

I think you should research the Turing test. It's a philosophical argument about what intelligence means. The LLMs weren't tuned to the test - they were tuned to intelligence.

And no - no software came anywhere close to passing it prior to LLMs. And yes, LLMs have now passed it.

Hello??? You there? You can't say "you aren't a very good programmer" and just walk away.

Well, you can. But it's cowardly.

Dangit! I was betting on -183.

Well - more importantly - with the absurd speed you can iterate insanely fast. It's amazing what you can do when you throw 1000 average programmers at a problem.

People who don't use coding agents really should stop making fools of themselves. It's painful to watch.