Np, search is getting terrible. Thanks for looking!

Source appreciated? Was this inside the research paper?

Just noticing: there’s 0 evidence in article that anyone is doing this. I just don’t buy that this is happening enough to matter. Interesting as interpretability research at best

That’s also precedent, and a template for using on institutions to break copyright. Still seems like good news to me.

Precedent is, in effect, new law and it absolutely does change who gets taken to court and the costs of defending your case. So, depending on which arguments the court accepts, I won’t need fancy lawyer. And it won’t require nearly the risk, creativity, or time that it requires of Meta’s legal reps today. Look at civil rights or environmental protections case law; big profile early cases were horrifically costly, and now compliance by company’s is largely by default.

Horrible people and companies can set good precedent, often without intending to. For example, plenty of criminals set and clarified due process law. So we absolutely could all benefit from Meta’s bad intentions.

We benefit from institutions that will be training their own AI, hosting data publicly, and have the resources to mirror a precedent. Care to cite sources that the arguments being accepted are going to carve out Mark Zuckerberg by name as the one person who can ignore copyright? I haven’t read the fillings, but this should be easy.

I read this as setting precedent that others couldn’t. Court cases like this are one way to make it possible for everyone to break an absurd law.

Worth remembering that any group could make a company. They are work, but not particularly class locked.

… Don’t they take a cut of most subs?

Also improves Teams/slows the enshitifcation. It’s harder to make the product bad when it’s hardly a monopoly.

The people who have made that category error aren’t reading this discussion, so literally reaching them isn’t on the table and doesn’t make sense for this discussion. Presumably we’re concerned about people who will soon make that jump? I also don’t think that making this distinction helps them very much.

If I’m already having the ‘this is a person’ reaction, I think the takes in this thread are much too shallow (and, if I squint, patterned after school-yard bullying) to help update in the other way. Almost all of them are themselves lazy metaphors. “An LLM is a person because its an agent” and “An LLM isn’t a person because it repeats things others have said” seem equally shallow and unconvincing to me. If anything, you’ll get folks being defensive about it, downvoted, and then leaving this community of mostly people for a more bot filled one.

I don’t get think this is good strategy. People falling for bots are unlikely to have interactions with people here, and if they are the ugliness is likely to increase bot use imo.

I think this confuses the ‘it’s a person’ metaphor with the ‘it wants something’ metaphor, and the two are meaningfully distinct. The use of agent here in this thread is not in the sense of “it is my friend and deserves a luxury bath”, it’s in the sense of “this is a hard to predict system performing tasks to optimize something”.

It’s the kind of metaphor we’ve allowed in scientific teaching and discourse for centuries (think: “gravity wants all master smashed together”). I think it’s use is correct here.

We attribute agency to many many systems that are not intelligent. In this metaphorical sense, agency just requires taking actions to achieve a goal. It was given a goal: raise money for charity via doing acts of kindness. It chose an (unexpected!) action to do it.

Overactive agency metaphors really aren’t the problem here. Surely we can do better than backlash at the backlash.

Taken broadly; literal management might be correctly optimizing shareholder returns for next quarter (cut costs at all costs), as the incentives encourage. The goal is no longer to keep having a business next year.

Directionally correct, but it does require self hosted agentic models that can compete with the automation running on corporate side. This is not obvious. It will be a new equilibria; maybe just a few more hours of poorly done work by a handful of consumers is enough to break some monopolies. Or maybe everyone will be attached to OpenAI compute, and we’ve just gained a new middleman for most interactions.

Then you should be able to easily give criticisms.

Major power computations I think are harder. I agree the US is falling, but I don’t think I’ve seen a serious analysis claiming that the US will become merely a power amongst a few dozen others anytime in the next few decades? We’re about to become much poorer and less diplomatically influential, but still nukes, still the huge military spending, still capitalists with hands in other economies?

And I guess I’m specifically reporting on my circle of american’s, who are both aware that we’ve burned a lot of goodwill and trust (though tbh, I’d hoped the trust was lost already after trump 1…).

The checks and balances held for Trump 1 relatively, and held for previous abuses before that (at least, sufficiently that folks would let the US say such things). I don’t think it’s obvious that they are irrepairable/irreplaceable: we could have a revolution and rebuild from scratch, as an extreme example. It is obvious that systems must change to do so; reorganizing the supreme court, changing campaign finance, etc. If they change, and how much, idk what to expect. But I think ~half the country knows it’ll take serious reforms, and it still wont put the US back where it was. Trust != systems.

At least in my circles, I think we’re aware? People are looking real hard at ways to leave. We’ve also got a higher than usual chance of reforming/refurbishing some of those broken guard rails. Fingers crossed.

The folks supporting Trump, on the other hand, already believed that these relationships were dead and bad. They’ll scapegoat somebody else for the decay; I do not see the avenue for this to be a learning experience.

It is very interesting to me that we don’t make this requirement for all large power users - factories, big suburbs, etc. Because we give power companies a monopoly (but don’t put them under state control), we often let big building projects force them to expand infrastructure (and then sell access as they do). So this is a whole weird thing with capitalism meeting very regulated monopolies, in a thousand different systems cause every local has different rules.

The thing that’s breaking our systems here isn’t that datacenters are big power users. It is that they can be built so quickly.

I’m surprised we didn’t make ‘bring your own power’ a rule before; I guess it’s infrastructure that generally is useful for many people to timeshare, and often isn’t fully used by just one party? Factories turn off some nights, for eg. And maybe it would be bad to have multiple power providers independently pumping power out?

If you read the article, it’s because power companies are monopolies and so we’ve regulated them rather harshly. They are often compelled to build infrastructure to meet demand, for example. We don’t make the provider of a steel mill, housing builder, etc pay (generally).

And that’s weird, right? It’s one area of the market where we do a planned economy, and all states manage it differently. Now it’s being stress tested in a new way.