I would say it's released in the spirit of open source. "Peer review" in the narrow sense exists primarily to assign prestige in academia; but there's nothing stopping anyone from "peer reviewing" the GitHub repository.
As far as I understand it, nobody is disputing the correctness of the Lean proof, or that it proves the conjecture it actually claims to prove. That's sufficient to consider the problem "solved". The natural language proof is a "nice to have".
That's an empirical claim. I could equally well say that doing an automated search of the problem space and having a database of results and open problems will identify vastly more interesting and valuable areas. Again, the idea that math is some exhaustible material is a metaphor, not an established fact. I'm willing to change my view as new evidence comes in, but I think we're going to have to wait and see what the landscape looks like in a few years.
As far as I understand, even with "publish or perish", peer reviewers decide what counts as an important enough paper to be published in a prestigous journal, and committees of peers decide whether or not, say, an expository article on arXiv or a textbook counts toward hiring or tenure. Again, as far as I understand, those things have largely not been rewarded in the past.
I like your marathon example, but maybe not for the reasons you intended. The community of marathoners decides the rules of a marathon. You don't need a hypothetical teleporter; you're already not allowed to use a bicycle, performance-enhancing drugs, or shoes that don't fit the specifications. The rules are updated to adapt to changing technology. Yes, I'm arguing that the strip-mining analogy doesn't make sense because mathematics is in the same situation. There's nothing stopping peer reviewers and hiring/tenure committees from changing the rules about which kinds of effort confer recognition and career advancement.
The problem isn't effort, though. All of the things I mentioned constitute effort and could be rewarded. The job economy was created by mathematicians incentivizing the proof of difficult theorems above all else and valuing all other work at approximately zero as far as career advancement was concerned. Now they're pulling a 180 and claiming that math was never really about proving theorems, but that's contradicted by their revealed preferences. The strip-mining problem only exists if they continue with the status quo ante.
As you point out, a model is not a person, so your second argument invalidates your first sentence. We can't assume that they're the same thing; that's for the courts to decide. It ultimately hinges on whether or not the courts consider a given use of copyrighted material as "transformative" or otherwise constituting fair use under copyright law.
I don't think the strip mining metaphor is appropriate. Mining is a zero-sum game; if I mine something, nobody else can go and mine the same resources I did. Mathematical problems don't go away when AI finds a Lean proof. They create new opportunities for humans to study the solutions, learn new techniques from them, identify promising directions for future research, discover alternative/more beautiful proofs, and write expositions for other humans.
Is there still a market for physical media? I assume they're transitioning to download-only because physical media are expensive to produce and most customers prefer downloads anyway.
You're confusing the weights of a model and internal chain-of-thought with the output of the model. Yes, we don't know a lot about how the internal mechanisms work. But with the correct prompt, agents will produce a worklog that documents exactly what solutions were tried and how the result was obtained.
Obviously, I'm not talking about the verbatim quotes I provided. There has to be some level of evidence that proves intent to commit a crime and the second quote is meant to represent that whole class of statements, but it depends on context. Any given quote won't constitute evidence in every case, but it will in the cases where it proves intent beyond a reasonable doubt.
I'm not sure why you think my argument holds no water when there are clear legal precedents that speech is not protected in some cases where there is "imminent lawless action".
I agree up to a certain point, but there has to be some legal boundary between freedom of thought/speech and literally planning a crime. I'm protected under the First Amendment to say "someday I'll rob a bank" but not necessarily "I'll rob this bank on Friday and here's how I plan to do it".
Right now, I would be willing to pay ~$20 extra per month for strong privacy, assuming the same capabilities as Opus 5.5. I don't see local models making sense economically any time soon unless you value privacy at $1000's per month or are fine with much lower performance on hard tasks.
You're completely right. I've internalized this on the technical side, but I still find it puzzling politically. Surveillance has been a salient issue in U.S. politics for almost 100 years now, since the invention of the telephone. I would like to think that most people are at least vaguely aware of Watergate, the Patriot Act, and Snowden. Then again, I should probably stop assuming that people know basic history, given the current state of education.
>I guess we probably want our tech to work exactly like this.
Do we, though? What if someone started an AI company that uses end-to-end encryption to make it impossible for anyone but you to access your data? Personally, I would switch to it in a heartbeat assuming it's competitive with the other products. I don't think it's the tech companies' job to surveil the population and prevent crimes. That said, I'm not necessarily against Anthropic or other companies reporting suspicious activity if their existing systems are detecting it. I'm just not sure we want every product to be forced into that data model.
Your follow-up about purely private tech seems to contradict your first statement. We can either have privacy or surveillance, not both.
I'm still puzzled that people's default assumption isn't that somebody is reading all of their internet communications. Ever since the Snowden revelations of 2013 I've more or less assumed that everything I type into a computer is stored in a government database somewhere. Not that I actually believe it is 100% of the time; there's a spectrum of trust, so I'm more confident that local apps on my Linux desktop are secure, somewhat confident in the end-to-end encryption of certain apps on my iPhone, but all bets are off for non-E2E encrypted data going across the internet.
I'm not sure what you mean by "correlation" in this context. The things I mentioned are strong evidence of causation. CO2 in the atmosphere causes warming by well-understood physical proceses, and human activities account for almost all of the warming above pre-industrial levels. In fact, the net amount of warming undercounts how much we have caused because we have also released aerosols into the atmosphere which cause cooling.
As I said, have a conversation with a top AI model and try to convince it that there is no evidence of causation between CO2 and climate change. AI is better than me or any other human at pushing back and presenting the right arguments and evidence for a specific line of questioning.
You can always have a conversation with AI about it. Relevant topics: physics of the enhanced greenhouse effect, IPCC methodology on attributing global warming to human activities, transient climate response to cumulative CO2 emissions.
I agree with your first point but not the analogy. Climate change is causally related to carbon emissions; if we had 100% carbon-free energy, there would be no concern with waste beyond inefficiency. Data centers can increase electricity/water prices, but it's not a linear relationship like climate change. Prices can stay constant or even go down if we increase supply by building more infrastructure, or decrease demand by limiting how much non-residential buildings can pull from the network. But yes, the conversation about utilities is ultimately more of a proxy for political frustration. If data centers became 10x more efficient, the same people would find something new to complain about because they're fundamentally against the technology itself.
I'm not against local governments setting zoning laws and regulations if they're concerned about the pace of growth. Personally, I'm not concerned because I think we can and should invest in infrastructure to keep up with demand. This will benefit all industries, not just AI. My point is that I suspect (I'll admit that I haven't done any research on this) that people wouldn't be against facilities with the exact same footprint if they didn't carry the label "data center" and this is primarily driven by a backlash against AI rather than the resource use itself. In that sense, they're trying to rebrand a political/culture war issue as a zoning or utility issue.
>Get out of here. Either you have a simple mind or you are commenting in bad faith.
I find it more and more common that people who disagree online respond by attacking the commenter personally instead of addressing their arguments. Saying "it's bigger" isn't enough; you need to explain why a building full of computers is fine if it's 100 acres but not if it's 1000 acres. You also need to explain why we shouldn't oppose comparable scale manufacturing, research, or warehousing facilities.
Cameras are different than data centers. Trying to connect the two is a rhetorical technique, not an argument.
The concern around data centers looks a lot like a moral panic. They've been around for decades and nobody cared before the backlash against AI. They're like any other industrial building, subject to the same resource constraints and zoning laws. I suspect if the owners of that land built a steel mill, a semiconductor fab, or anything other than a data center with similar power and water requirements, nobody would care.
Obviously the scale of retrieval from memory is different, I'm not disputing that. The empirical question I was referring to is what percent of its training set the model is capable of reproducing. What looks like "large swaths" of text to us could be less than one in a million for all we know.
I'm skeptical that there is a single spectrum like you're describing. It's not well defined. Say a human and an LLM prove a new theorem independently (without external help, i.e. from their own neural weights and reasoning). How do we measure how much each of them copied from previous work, as opposed to having learned from or been influenced by it?
Humans are also capable of memorizing long texts and sequences of numbers. It's more a question of scale than a fundamental difference. The point is that LLMs are like human brains in the sense that they're not databases that store text, even though they can and do memorize things.
Just because a model can reproduce parts of its training set doesn't mean that's what it's doing when it solves a programming problem. It can also reason about the problem, draw on its knowledge of algorithms and data structures, write tests targeting APIs it's never seen before, generate synthetic data and run experiments, etc. etc. Also, the claim that it can reproduce large swaths of its training data verbatim is an empirical one. I would be surprised if it could even reproduce 0.1% of the books it's ingested, for example.
Saying that AI can't do anything but copy or steal from humans seems to be a rhetorical technique used by people who are still unaware or in denial about the capabilities of the agentic systems released in the past few months. They can now one-shot theorems and programming problems in a few minutes that would be difficult and time-intensive for even the 99.9th percentile human expert.
I don't agree that AI tools will have the capability to escape any sandbox. It's a question of convenience and engineering effort. For example, you could design systems that run under a formally verified hypervisor in a physically secure airgapped network. Companies aren't doing that because there aren't currently any incentives to do it.
It didn't copy any source code from any external projects. I had it write a stratified sampling renderer for ground truth, then had it implement feature by feature by matching the pixels. Unless you mean it "copied" it in the sense of third-party code being part of the training data. I don't think that definition of "copy" makes any sense given how these models represent embeddings. It would also imply that humans are "copying" the things they've learned from.
I've been a vibe-coding skeptic for years, but because of the math breakthroughs of the past few weeks I decided to experiment with the latest models on some test projects. They're a lot more capable than I thought they would be. I agree that it's easy to create an unrecoverable mess, especially when you're one-shotting a lot of features without detailed instructions. But I find that as long as I'm strict about the API boundaries and force the agent to work in small chunks, it's pretty effective. As one example, I got it to write an SVG renderer in a few hours (not the whole spec, but most of the path features and text rendering), which would have taken me at least a week just for the coding part, plus extra time to learn the algorithms.