Ah I got excited for a second thinking this was some kind of AI test on a Warhammer 40K game:)
2,786 karma · joined February 28, 2020
jon@vicena.ai @OerstedIV
But indeed, if Pangram is marketing itself as "near-zero false-positives" now, that entails that the are aggressively biasing it towards being conservative, likely leading to lots of false-negatives and low match scores on AI content.
But the analysis of the internal admin pages for generating articles, the API and the interview structure are pretty damming evidence that this is largely driven by AI.
Their revenue has always been sustained by the fact that their technology needs to be constantly replaced because it keeps getting better. The moment it stops getting better, the replacement rates plummet and so do their revenues. It's also really not that hard to compete with them when they get complacent.
I'm not counting Nvidia because they don't produce semiconductors themselves, they are a different kind of business.
They are indeed an example of those that cater to consumers and/or build popular products based on foundational tech from others, like Apple or Sony, which do tend to be quite profitable.
But the actual deep-tech semiconductor firms? They may be critical to the world economy, but they don't actually make that much money comparatively. In many cases they are not real monopolies, it's just that no one else wanted to continue investing in a shitty business model. Only the likes of TSMC and Samsung were okay in playing the low-margin game, but most US players left the board.
I believe AI has a lot of the same characteristics.
My perception is that there's a cultural issue suppressing the investigation of the causes of quantum mechanics, making it much harder for a better theory to flourish.
It's also really-really hard. If you take my earlier example, the probabilistic theory of the coin-flip is worlds apart in terms of complexity from a real understanding of the forces acting on the coin when it's flipped. It's a massive gap to jump over, you have to come at it from another easier and longer path, like it was done with classical physics.
We have seen plenty of examples in other industries where you can never really stop investing a ton on R&D with diminishing returns (like in semiconductors or pharma), because the moment you stop newcomers overtake you.
Or the whole thing becomes a commodity with lots of competitors, where technological advantage is overtaken by marketing as the dominant force.
Or you really are left as the only player alive, but you realise that the market cannot absorb higher prices for your product by then, they prefer just not to buy it. Perhaps you are the only player alive because the business has become so low-margin that everyone else has abandoned it intentionally.
That Silicon Valley pitch you are echoing rarely works out as advertised, even for the winners.
The R&D expenditure is a critical requirement for the inference profits, to the point where we should probably lump their financials together, at which point is definitely not profitable.
What will it look like when R&D plateaus (and yes it definitely will, but it could take a while), investment falls, and a few main competitors remain in the music chairs?
It's very difficult to predict. The inference profits we are seeing the profits of a company that is temporarily ahead, but the revenue will level-out in a more stable market, depending on how many survived. It's also hard to tell where the costs will be at the end of the game, with constant efficiency optimisation mixed with cost increases for higher intelligence.
I think it will be quite similar to the semiconductor industry, where, yes there are some key monopolies, but they are not the initial big players, and none of it is actually very profitable; while the real profits are reaped by those that make popular consumer products based on the foundational tech. I guess the main difference is that OpenAI and specially Anthropic have been quite effective at directly tapping into the consumer market rather than remaining technology providers.
I believe the point is that the Standard Model is still built on Quantum Field Theory, but looking into what's really going on under quantum mechanics is a bit taboo and very few people are spending time on it.
Frankly, I acknowledge that I am an outsider and that surely my perspective is far too naive and simplistic, I'm sure the reality in the field is far more nuanced. But still, I'm in good company when I question what's going on, like Einstein and many other prominent physicists have.
It is true that the end of the conversation was just "shut up and calculate", which is wise on some level, but it also implies that the mathematical model is the end-truth and you should unquestionably just build on it. There is a consensus that the mathematical model of quantum theory is really what is mechanistically happening, and that the wave function is a real physical thing.
But one can also argue that it is merely a probabilistic theory that, yes, describes the probability distributions of what will happen extremely accurately, but still is a partial answer and it doesn't actually predict what will happen. In every other context, we use probability to describe processes that we don't have a complete model for. It is a bit presumptuous to think that in quantum mechanics probabilities are somehow fundamental and not emerging from more deterministic underlying phenomena.
You can have a theory that says that there's an equal likelyhood of getting heads or tails on a coin, and that theory can reflect the statistics extremely accurately. But that is obviously not the end of the story, there are deeper mechanics determining if you get heads or tails at each given time, it is not fundamentally random, you just don't fully understand what's going on yet.
My naive intuition is that this is mostly just physicist publishing papers and throwing some exotic ideas out (we still don't know enough about dark matter), and that the black-hole model will remain as the consensus. But it's not as unquestionable as you say it is.
> An alternative to the black hole (BH) scenario has been recently proposed in terms of a supermassive compact object composed of self-gravitating fermionic dark matter (DM).
https://academic.oup.com/mnras/article/546/1/staf1854/843111...
> Sagittarius A* Might Not Be a Black Hole
Games make significantly more money than movies and TV now. More than 65M people bought Skyrim, and many more millions would recognise Fus-Ro-Dah just from spending time on the Internet anywhere near to gaming circles.
The defendant could have told their lawyer and it would have been trivial to research and demonstrate. Not sure what went wrong there.
It is the name of the headline magic ability in Skyrim, which is probably among the top ten most popular videogames of the last 25 years.
It is written like a play where characters with real personalities have fierce and witty debates. It is rather fun!
Seveneves is probably my favourite, I guess I prefer harder scifi. Fall; or, Dodge in Hell and The Rise and Fall of D.O.D.O. were pretty good too, although less so. I want to like the Baroque Cycle but I haven't managed to get into it so far.
But it can still not run truly capable coding agents at reasonable speeds right?
That doesn't make sense for the cost of quite a nice new car. If you want to spend that much, build yourself a proper workstation at least. Clearly a computer with the form-factor of an Apple TV is not going to be ideal for getting the highest performance for your buck.
Is it just wannabe vibentrepreneurs with too much spare money hyping each other up?
Not sure what that means for the metaphor, but there you go.
The lowered production costs enabled by AI should to a degree insulate you from the usual financial risks, and should allow you to be more authentic. It might seem risky to make niche content, but I think it could help attract a real initial audience, never underestimate the demand for true authenticity.
It might sound strange talking about authenticity when it comes to AI, but I do agree with your perspective, that it can empower creative thinkers to produce content with real authorial intent directly, and without the compromises inherent to large production teams.
You seem to be operating like a corporate production company with large costs that needs to appeal to the lowest common denominator to get ROI, but you are doing it without the benefits of the higher production budgets.
You are wasting your advantage. Do the things they cannot afford to do because they are too risky or too niche. If you say that AI allows you to empower your writers, don't waste their talents on brainless content!
EDIT: Indeed looking at the overall rankings for text again, the list is rather strange, a lot more about writing style than intelligence.
Look it up, it’s a bit difficult to explain concisely in words but it is intuitive visually.
If we are thinking of intelligence and price, a model will be in the Pareto Frontier if there’s no cheaper model of the same or higher intelligence. Or if there’s no more intelligent model for that price or lower.
EDIT: See this chart from Artificial Analysis: https://artificialanalysis.ai/#intelligence-comparison-tabs
So for example DeepSeek V4 Pro can be considered a frontier model because there's no cheaper model that is as intelligent.
For any solution in the Pareto Frontier, there no "no-brainer" alternative, in the sense that there's no other option that is better in some way without giving up something else. It's the best of its "weight-class".
> I think you should spin up a whole bunch more instances, and try to cause an integer overflow so they they owe you $978 Trillion.
> How do you properly mix IO/compute (in any language)?
Good question, it’s a really hard problem.
In principle you are meant to use threads, the OS will automatically switch to other work while you are waiting for IO, like with async. But there are difficulties with threads too: memory overhead, switching overhead, thread limits… Although I am getting the feeling that they have a much worse reputation than they deserve. More projects should try switching to threads and actually measuring the difference.
Isn't it just like in async and any other cooperative concurrency model? At least in multi-core async, that message can be handled in a different core so it's not completely stuck. But sure the author doesn't like that work-stealing cost happening automatically outside of their control, fair enough.
Regarding mixed IO/compute and preemptive scheduling, well that's what threads are. They are not as lightweight, but they are there and they are quite ergonomic to use in Rust.
I could even imagine an async executor that simply starts a thread per task, so that you can keep the nice syntax, but that's obviously not ideal from a performance standpoint. Tokio already dispatches tasks to a thread pool, it is not completely cooperative, it's the sensible way to do this. And there's always tokio::spawn_blocking too.
PS: Actually, it is underrated how well designed classic threading is in Rust. It was just going out of vogue when they did it, but they addressed so much of the complexity that was around for years in Java, C++ and the like. They did actually mostly achieve the holy grail of fearless concurrency and with a more ergonomic design.
Is Project Tina a bit like the Actor Model, but having actors pinned to cores?
And I don't understand how Tina deals better with the problem of compute-heavy tasks blocking the thread. It looks to me like it is also cooperative concurrency per core, and if one Isolate runs for a long time the other Isolates in that core will not be able to handle their messages.
It's a bit like having a system composed of microservices (nanoservices?) that communicate via function calls.
It sounds a lot like the actor model, but I always found the classic architecture too limiting: requiring every actor to be a single-threaded message processor, instead of being able to handle requests concurrently. It's not too different from classic object-oriented design either, with singleton services.
In some project I've called my concurrent services Gods just to have a bit of fun with it :)
Async is just a way to describe a tree of concurrent tasks that may depend on (wait on) each other at certain points. It is mostly declarative.
Tokio has taken over as the default choice, but there's a reason why it's not part of the standard library, it is not meant to be the only choice.
It's connected to all papers of course, and all kinds of scientific simulators and specialised models. But I'm currently in Shanghai talking to labs to join a CloudLab (and hopefully setting up our own robotic labs), so that AI can actually order real physical experiments that are executed cheaply, efficiently and seamlessly as tool calls.
Through experiments like autoresearch we have seen that AI is already, if not always smarter, at least more systematic than humans at following the scientific method relentlessly (hypothesis-experiment loop). Let's see what we can do by connecting it to the real-world :)