1. That’s a legal grey area that could result in significant legal exposure for Anthropic.
2. Even if someone wants to do something that’s in a legal grey area, or more (hacking others, another item limited by Anthropic), is the idea that such people won’t have the ability to use AI to do those things? I don’t understand the criticism that just because one company might institute some guard rails that those who want to conduct legally grey or black activities will be unable to use other AIs to do so?
You might say that the legal grey area involved here is significantly smaller than the legal grey area involved in training these models on literally all of the copyrighted material they could find. The difference seems to me that it's more likely that Sony would pick a fight about this becoming a thing, whereas the training data is a horse that has already bolted with an unclear path to remedying it.
To a first approximation I agree that it’s similar to industrialization. However there’s a key factor which distinguishes this case, and which makes the analogy less than perfect: there isn’t a profit motive in mathematics, so there was no push for this development from inside the field. Instead, the industrialization was entirely exogenous, from companies which do in fact have that profit motive and view math as a means to achieve it.
There might be a “proof motive” in math which one could argue will drive the full automation of theorem proving. But we must recognize that profit is more universally valuable (i.e. it has value outside the industry in which it was created), whereas proofs are only valuable inside the math community (at least the pure math proofs the labs have been publishing thus far).
Do rich people like opera? Yes. Is that a profit motive? No... but it's a shibboleth for wanting to be rich. Do you get it?
It's ridiculous to say, no profit motive in mathematics. It's like saying, no profit motive in LISP. But cmon, you think the Collison brother who took Math 55 and liked LISP: I know that in a pedantic sense what you mean. But in terms of academic departments where people say they are no interested in money and it turns out they want to become billionaires... Math Department ranks pretty highly. Always has.
I agree that being technically gifted has been a great way to make boatloads of cash in the past few decades, and that this correlates with being good at math. But I don’t see why this implies that enjoying math means someone is more money-minded than others. If anything, the students in pure math departments are intentionally forgoing the potential cash conferred by their technical abilities in favor of the academic pursuit. In anticipation of a reply observing that math PhDs regularly go into quant, I counter that quant programs reach out to interview students in many other fields, and more to the point, spending half a decade studying algebraic number theory is surely not even close to an efficient way to build wealth.
In response to your second point: even if mathematicians desired wealth more than other professionals, how does this demonstrate the existence of some kind of profit motive in math?
I think it's more that the quirky neurodivergence makes it much easier to look the other way for the skeezy, and sometimes criminal, ways that banks make money.
How you get somewhere is important and tools change how you get there.
You can drive 1000 miles and arrive at the destination on the highway quickly.
Or you can take the slow road and walk. You discover unrelated things, make new relationships, see the scenery, listen and smell what is around you and have time to contemplate things which may or not be related.
This particular tool economy robs us of all these things in the rush to the destination.
I'd sacrifice "progress" a little for some of the better bits of humanity.
The thing is that if you don't learn to drive some destinations are unachievable. And destination is more important than the journey unless you travel for pleasure or personal development.
The first sentence is actually a much better jumping off point for the other side of the argument than yours, because the mistakes involved in the statement should set off alarm bells for people who've walked with mathematics.
What do you mean by "place"? What do you mean by "more"?
Are there more places as marked by real numbers between 0 and 1, or places as marked by integers? Could a machine that can count integers 1000000 times faster than people "visit" all those real places between 0 and 1? What would it mean to visit all the places between 0 and 1? Once the machine visited all of them how would it help people explore that territory?
What do the general ratios and rules between paths and areas mean about how much path it takes to cover an area?
The real world is quantized in way that math isn't, of course, but that's one way the analogy can lead you astray.
And even just keeping to the real world rather than math, it's not all about the sheer velocity with which you can cover an arbitrary distance metric. The primary bottleneck is less that velocity than the amount of time you have to direct to place scaled to your capacities. The car's velocity can let you trade some of that time for fast travel between farther places, and with a road network can arguably even increase a certain variety of places, but that's a very different statement than more places. And it's also a wall against attention to everything in between.
You can just say "I like driving my car" and "I like the idea that I can do things with AI that I don't understand" and "I like to repeat my argument rather than engaging substantial point & counterpoint."
That's clearer, and hey, if that's what you like, enjoy it.
And you still will not, in fact, be able to get to more places in your car, though you'll probably be able to continue to refuse to understand that or why it matters, and maybe that's just as good.
Is this about driving to an island? If you need a plane or boat to the island and still call that trip "walking", then someone can take a plane or boat to the island and rent a car and call it "driving" or "driving plus walking".
Or if you're just saying that a very good backpacker that happens to be in greenland can get more places by walking because of a lack of roads, that might be true. But for a normal person in a typical household, their walking range is not that big. If you're in the US or Europe the roads reach almost everywhere.
Not OP, but actually I do because I have been forced to live in places where it’s dangerous to walk and where all the scenic landscape has been replaced with asphalt.
I get your point in general, but I think vehicles are a poor analogy because in many ways they have changed the places we live for the worse (and yes I’ll concede occasionally for the better). It turns out we need places to park all of these vehicles and they compete for space with humans.
That’s very narrow-minded. I love hiking and biking and exploring new places, meeting new people. I also appreciate that I can use the highway to get to the destination conveniently. Often so that I can enjoy the very experience I’d miss according to you.
They build cities around the car, they defund public transit, they close down rail lines, they build new housing a 5 km away from your closest grocery store, and 20 km away from any cafés and other cultural venues. Your job is located 50 km north of where you live, your partner’s job is 60 km in the other direction, and your child’s school is 10 km in the third direction.
They neglect to build sidewalks, and force you to walk 1-2 meters from cars racing at 70 km/h. With all the noise and high risk stressing you up to the level you just wanna scream.
stevenhuang: No one forces you to drive. You can still walk.
Don't tell me you dream of a maoist city where only bike is allowed and "car-brains" (as anti-automobile fanatics call us) are considered "enemies of the people" right?
do you think, perhaps, that it might be possible to cool the discussion? that there might be some meaning located between the picture painted by the person you're replying to and your own?
or do you sincerely think that the only options available to us are "all cars banned" or the alternative posted by the person above?
So drive then. You can also move. The point is there are options. Many cities have strong infrastructure and have embraced walk ability. If not driving is such a big deal to you, why not do something about it?
You're arguing like an edge lord here. What he is saying is that you're effectively forced to drive. Not so difficult to understand. There are plenty of people fighting for change but because our political and economic landscape is captured by big business, it's not so easy.
Not everybody in the world is effectively forced to drive. Sure, maybe if someone has the misfortune to be an US city they have no other option but there are other places that are not like that and cities with good public transport and relatively low car usage.
Yes but GP is describing basically the norm, while I'd argue the minority here lives in a public transport utopia like the Netherlands or Denmark. People generally have very few options outside car ownership, even in cities.
It's pretty well documented they deliberately made things impractically large distances apart and scrapped the idea of footpaths, to encourage car sales, paid for by car companies to your local city government several decades ago.
Could you please stop posting unsubstantive comments and flamebait? You've unfortunately been doing it repeatedly. It's not what this site is for, and destroys what it is for.
Unfortunately, I cannot think of a single politician right now who is a thought leader on regulating technology, which is exactly the kind of person we should be looking for. People who understand that we don't need every website to give us an annoying popup saying that they are using tracking cookies but we do need the option to opt out of AI features for basic services.
What? Only websites that want to track you before opting in have cookie banners. HN doesn't have a cookie banner, in case you haven't noticed. I'd say your issue isn't with consumer protection, but with pushy marketing people (or worse). But they're good at marketing, hence the cookie banner nonsense complaints, which redirect the anger towards laws requiring that creeps announce their presence, rather than the creeps themselves.
Two possible walls for advancing math with AI: first, it could be that the kind of problems that AI solves are the ones that don't open new avenues to concepts and ideas, and second, problems complexity is usually not linear, and the new and interesting problems to solve could be at a level of complexity beyond what AI can currently solve. Perhaps mathematicians could work to find ways to help AI cope with those walls.
> My recent take on AI is that people haven't adapted to the new huge potential of the tool, and don't dream big enough (yet).
I guess you can dream big enough only if you have access to the really interesting models. The 20$ subscription won't get you there, and the 100$ one is already out of reach for so many people.
Yes, this. Unless there is a true singularity where no more questions are ever asked, there will be definition be infinitely many problems to solve, and most of them we have not even glimpsed yet.
Singularity? Probably not.
Tech bro dictatorship? Unfortunately a bit more likely.
None of the above? Then we have a small shot at a great future.
Having a machine that solves hard problems must be a good thing.
You get an AMD GPU, put it in an external case ("enclosure") connect it with usb-c or thunderbolt and that's what's being used in the app. For inference of course
But then how do you even make the driver install on the ipad ? Aren't drivers supposed to run under system-level privileges (which iPadOS will never let you run under)
>The drivers you build with DriverKit run in user space, rather than as kernel extensions, which improves system stability and security. You create your driver as an app extension and deliver it inside your existing app.
Totally disagree on the reevaluation of architecture best practices.
Agents are as prone as forgetting to update duplicated code as we are ( or almost ), because they work at the same level : text files spread everywhere that you have to open and read to know what's inside.
The fundamental reasons for those architecture best practices are still there, and it's actually quite convenient and also a good reason to remain optimistic in our ability to find a job as a professional software developer.
Once the rest of those talks are published on YouTube (usually in a couple of weeks), you’ll see that there were some other talks that were much better about conveying a nuance in a balance between the tension of “AI can do everything for me, let it write machine code” and a session titled something along the lines of “your code base is your prompt“.
One presentation, I think it was my favorite one, talked about how an LLM had a very difficult time understanding what happened in a class when the name of the class was poorly descriptive of that class behavior.
Good quality code still matters. The LLMs are based on language! Human language!
That's indeed a weird feeling when going back to another language after having coded in rust. First you're happy not having to write any "unsafe" keyword. Then you're horrified for the very same reason.
i'm surprised modular's doesn't get much traction. The promise seems super interesting, and chris latner has the record to back up his claims. If someone has an explanation..
There are none, this isn't a problem specific to OR as much as it is a problem with serving LLMs in general.
If you use any other meta-provider that routes your requests to third party providers, you'll likely face the same issues. If you try using any of those providers directly, you'll likely face some of the same issues as well, except you won't have the option of quickly swapping to a different one and taking your credits with you.
Extreme variance in quality and feature support per provider is probably the biggest obstacle holding back adoption of open weights models.
I've had opencode go + opencode work reliably, though I'm skeptical of how robust their data security claims are in practice because they suddenly blocked accessing Deepseek unless you were okay with the data going to China where true data privacy for something like that is illegal, which makes me wonder where it went before, which weakens my trust. It's also a lot less useful now that Deepseek is so much more expensive.
Openrouter is useful for quickly testing various models with just one API. In development, it's useful. I would not run it in production tho' for all the caveat mentioned. Go to the first party provider directly, it's cheaper usually. And the cost to rewrite to use their API is usually noting (you can even have both and a feature flag), especially if you just vibe code it.
I've been using Merge AI Gateway and it's been useful so far. They tend to add new models quickly, and support has been responsive. https://gateway.merge.dev/
I was looking for an LLM gateway and saw that the most popular one had just had a massive supply chain attack, so I wrote my own. Took about 2 weeks and initially I wrote it as a provider for pi coding agent. I connect to moonshot, qwen, Gemini, zhipu, anthropic, deepseek and OpenAI. I use models.dev to load model and pricing info. Adding new providers is pretty easy because I have a standard internal format and each provider has an adapter that translates between my standard format and that required by the provider.
Fireworks has high variance amongst models and while some are served correctly; many of them are junk / broken and degraded and it seems like they don’t even know; because even running 1k MMLU Pro questions would flag it very quickly.
I've used Requesty (https://www.requesty.ai), let's you pin down providers and build your own routing policy so you at least somewhat know what to expect.
Just use a single vendor. Literally nothing wrong with that, and you avoid the complexity of both n-1 of the vendors (leaving you with the highest quality vendor) as well as the issues with the aggregating layer.
Not sure why people are drawn to this particular blunder. The promise of vendor neutrality maybe? I'll take working product over vendor-neutral slop anyways.
They used a closed list of 3 vendors in prioritized order, and got 429ed out of two of them, while the third one stopped serving the mode.
This is less of a problem if you're running an agent locally and routing your problem to OpenRouter - you can pin to one or two models for consistency and just switch models when something goes bad. But the article is specifically about production traffic.
Never before have I heard this sentiment, NGL. Vendor-neutrality has been an OS(/FLOSS) darling for, well, the whole time.
RE:"single vendor", if this post is to believed then you might have picked one that has 100x the tool calling errors for the next SoTA model, if your single vendor serves the next SoTA in the first place. It also completely erases the notion of competition driving down prices -- that would only hurt you in the short term, but obviously would ruin the whole ecosystem long term.
How does it erase the notion of competition driving down prices? Endpoints are largely compatible, so the code change required to switch from one to another is trivial. Don't load 6 months' worth of credit in an account, keep it tight. There's fairly little lock-in.
The most significant lock-in to me isn't even something you mentioned, but rather it's model related; I personally put a little time into trying to optimize my prompts every time I change models, as they all have their own unique... flavor.
As for tool calling errors, it seems like first party providers are among the best, I got the feeling that's what he was suggesting, though of course that's why you test. You can also go directly to Together.ai or whoever else you please.
Like others have said I think Openrouter seems neat for testing, but even just as a hobbyist I've been drawn to go direct to particular providers due to irritating little issues that I now see just weren't me.
>Endpoints are largely compatible, so the code change required to switch from one to another is trivial.
Wrapping a specific implementation in a neutral function is something you learn to do in year 1 of programming.
This specific issue and argument I see in lots of different aggregator dependencies, Terraform, LiteLLM/OpenRouter.
They promise to save some hypothetical work in the future if your boss asks to change vendors, and it turns out to be very trivial work that is just a regular part of our programming job, changing a couple of lines in order to change vendor.
It's worth noting that there exists a similar set of technologies with a reasonable tradeoff, using a framework that targets different user-platforms makes sense, write-once and deploy at iOS and Android is a reasonable tradeoff, but because you are deploying to those providers simultaneously and it's a user-choice so you don't get to pick one or the other (without losing clients), there's still arguments to chosing just one and losing market share, or doubling the workload and building native for both, but this is a true engineering choice. I feel like stuff like OpenRouter and TerraForm take elements of these frontend abstraction technologies and wastefully apply them to backend tech.
A particularly egregious case is when there's an aggregation layer for aggregation layers, say, a tool that generates TerraForm or Chef configs, or a tool that generates Docker and Podman containers, or a tool that generates LiteLLM/OpenRouter configs. Sounds dumb, but it happens when there's a market share for it. Can even get to 3 layers deep.
At the foundation might be an aversion to making an irreversible choice, which is an innate emergent psychological phenomenon, but is supported by the Bezos Amazon policy of reversible and irreversible doors. But again, even if you want to be light, using some of these aggregating tools isn't necessary, you can just build on top of a tech, and switch later. The only thing you get with an aggregating layer is that the API ends up being the common denominator so you lose out on the competitive advantages of each choice, or are forced to use even more complex API logic like LLM(commonParam1, commonParam2, vendorParams= {"vendor1"=:{"vendorParam1":"blabla"}} or worse, use hard-coded aggregator provided mappings between the aggregator API and the vendor API that may be incomplete and relies on updates from the aggregator dev.
I think OpenRouter’s value proposition is less about avoiding two hour developer tasks and more about having a single place to establish policy controls and dynamic selection based on current pricing and performance data. If you can’t actually do that - for the reasons described in the article you end up pinning - they can’t deliver. But it seems to work for some use cases.
Still the best way to ensure such policy control is to have 1 provider, tops 2 or 3.
Having a router thing that reroutes to 18 different vendors is of course no way to ensure any policy control, you can add all the internal buttons and dials on policy control and ISO and GDPR compliance, but all it will do is (incorrectly) check compliance box and increase compliance risk to the 18 different vendors.
In practice most openrouter users look for the cheapest vendor, and they tend to go for chinese vendors, who love to price dump and don't have the same views on contracts and IP as the west.
Who is this vendor that is consistently providing high quality inference for all families of open weight models at a competitive cost?
That's a serious questions that I really interested in the answer too. I have 25 providers included into my coding agent harness not because I care about vendor neutrality, but because I have to keep adding new ones as inference quality degrades at the providers I was using. Its quite tiresome.
it's also my opinion. Which makes me wonder : isn't there an opportunity to create a variant of rust that would make absolutely zero compromise on UB and safety at the detriment of user experience ( which we don't care about now, with AI generating the code) ?
Like programming in a kind of super strict IL. Or the opposite : super poweful, super abstract language, yet extremely strict.
But this is already happening with iPhones. Apple is touting on-device AI and only the latest phones offer the full capabilities. Newer phones will be able to run better models, so the incentive is there as soon as someone makes the killer app that only makes sense when the model is running locally on your phone.
My recent take on AI is that people haven't adapted to the new huge potential of the tool, and don't dream big enough (yet).
It's like seeing a car engine for the first time, lamenting about the future of horses, and not realizing it'll also help us fly.
reply