The more I read HN the more I realize that I am more of an empiricist than a rationalist.
So many people here are opining as if they are correct. I have a large amount of doubt and I like the fact that the world is structured so that people can try things out and see if it works.
If the world was run by the naysayers in this thread, almost nothing would ever get done.
I have doubt but I don't believe in the idea of something like this being strictly impossible.
This is turning out just like everything else. We establish a certain conclusion that is preferred and then walk backward into whatever evidence supports that conclusion.
In this case, AI needs to fail for certain groups. It is an apparent loss of identity. The notion of it going to space where it's a lot harder to get at (protest/regulate/etc) must really bother these people. I don't need datacenters in orbit to succeed. Whether they do or do not doesn't bother me very much. The concept is interesting but not essential. The inverse is clearly not true with the counterparties. I really don't understand why you would go on the internet and be a total shitlord regarding how orbital datacenters are impossible and fuck you for even suggesting they might be.
Oh my god, it'd be way easier to, I don't know, build datacenters near Tashkent, Uzbekistan, where [0] is located. Here's your almost unlimited solar power, cheap land, and still better latency and easier cooling.
> it'd be way easier to, I don't know, build datacenters near Tashkent, Uzbekistan, where [0] is located.
To build? Yes, it would be easy. But to not get those datacenters annexed by Russia would be pretty hard. Russia experiences shortages of hardware right now, I'm sure Putin would be happy if we placed some ready-made datacenters nearby so he could grab them.
Uzbekistan doesn't even have a common border with Russia; it borders Kazakhstan instead. And even Putin can't teleport over a country, grab the datacenters, and teleport back to Kremlin.
> Uzbekistan doesn't even have a common border with Russia; it borders Kazakhstan instead.
I don't think that it matters. Uzbekistan is highly dependent on Russia, Putin probably wouldn't even need to cross the border, he would need to ask Uzbekistan to nationalize a datacenter, and then some Putin's relative would buy it from Uzbekistan.
Are you ready to invest in a datacenter to see would it work in practice?
Putin is dead within 10 years, he doesn’t matter (unless he uses a nuke). Time is running out not just for him but for Russia if they don’t get their act together. Within the next 20 years in the same, may apply to the United States in certain respects the turnaround across the board needs to happen fast after the present clown show in the White House. right now, what are the odds thru the end of this century? China, Russia, United States, Western Europe.
> Putin is dead within 10 years, he doesn’t matter (unless he uses a nuke).
The question still stands: are you ready to bet your money on Putin's death? Do you think that Putin's death would change Russia's behavior enough so your datacenter in Uzbekistan wouldn't be in danger?
Not really, no. It exports more to the UK and Switzerland than to Russia; it imports more from China than from Russia. Sure, Russia is Uzbekistan's major trade partner, but Ukraine had Russia as an even bigger trade partner — and it didn't make Ukraine cave in to any of Putin's demands, if you haven't noticed. So neither would the Uzbeks.
Sadly this world is being run by masters of hype that appeal to the scientifically illiterate. Their super power is getting politicians and fund managers to support techno-religious non-sense with other people's money.
What's the goal or even the story at this point ? More compute will cure cancer and make those humanoid robots finally do something useful ? Then why are we letting millions of consumers use compute to create slop? Why is a significant part of the training data pornographic materials ? If we are so desperate for more compute it needs to in outer-space then why use it to remake the Odyssey ?
I really didn't like Opus 5 and was vocal about it. Opus 5.5 is noticeably better, but literally just yesterday I was fighting parsing some summary it had given me and it was a bit like the old days. IOW, it is better but still has some room to improve.
Yay! New model day. The most likely time the model isn't nerfed and we get a glimpse at the latest capability.
I wonder how long this kind of excitement at new models will last? I mean, now when I get a new iPhone I don't get the same rush I used to. So I wonder if this hype feeling will fade as we plateau in model capability.
Considering Fable 5.2 is not even worth using over Opus 5.5, can we put the model nerfing nonsense people put all over the comment section to rest now?
If this turns out to be accurate, I think that would be a pretty unethical thing for Anthropic. Even if it is just a "we had to allocate compute to this new model release".
It should be fraud to swap out the model. We need a term for this, something like a bait-and-switch.
> "We need a term for this, something like a bait-and-switch."
The media coined the term "dual quality" for a European food scandal a few years back, in which conglomerates (including Coca-Cola) adulterated products with cheap, substitute ingredients, if they were destined for distribution in specific markets (mainly, lower-income).
The packaging exteriors were the same. You wouldn't know which version you got unless you were paying close attention to taste. It was essentially what the software industry calls "A/B testing", except with sneaking things into what people eat.
I don't have any good reading links on hand; here's one source,
What people claim is not "some experiments degrade performance", it's "Anthropic nerfed the models [because they are evil and do not like the user etc]".
Like most experiments do not save the company money, there was an experiment to test the new fable, people got to use it at opus-5.5 usage. But this experiment clears the thinking blocks when people go back to the vanilla model; not sure if Anthropic overlooked this, but idk, it does not sound like "nerf on purpose".
There are experiments to adjust the effort level, at the moment the ones that I've seen switch to a higher effort. Again this is not a nerf; might be annoying perhaps, the model does start to behave differently (and consume more tokens). But it's not a nerf.
Then there is apparently one experiment that replaces opus-5.5 system prompt with the one from opus-4.8; the model is the same; this experiment is perhaps a bit more questionable, but still it's not "anthropic nerfed the models".
If people think they are getting nerfed they should check what experiments are enabled (claude.json and session logs).
I'm pretty convinced that those who have constantly been posting these complaints, probably for every single model release out there, are indeed wrong.
> And not just in this instance, nerfing has been complained about in a cyclical fashion for many months.
No, it gets complained about constantly even if there is no model coming out. A completely unrelated thread on Claude will feature comments like "can I just say that [model] has suddenly become useless this week?" Merely asking that question is bound to get people agreeing with you because we are evaluating all of this on vibes and placebos.
I too feel that way sometimes, I don't think it's ridiculous that Anthropic might silently downgrade you as demand warrants. But asking others if they "feel" something I feel is less than useless.
I believe fraud is a sufficient term. You claim to be selling X but you actually sold them Y. Bait and switch is a selling tactic to avoid fraud where they make you believe product is X and then when you go to buy they tell you X isnt available but they have Y. While also unethical, it is not fraud because they did tell you what it is you are buying before you buy it.
if they just straight up dont sell you something you bought then its fraud.
I think there is a ripe opportunity for expanding the definition of fraud to include changing the "ingredients" of a named product on the market. It is not limited here, we see it with food, and even in the medical field
(I will die on the hill that medical fraud was committed when mRNA therapy was used as a treatment for covid and it was 'sold' to the public as a "vaccine" despite not being the same thing as any vaccine before it. The definition of vaccine was expanded to include it without sifficient public disclosure to warrant informed consent imo. This is not a statement about conspiracy or the efficacy of vaccines or a moral judgement on whether it was for the greater good or not.. it was simply an act of fraud that robbed the public of proper informed consent by making people believe it was X when it was really Y.)
So yes, I think there is a large need for the public to at the very least put a strong handle on the ability to change what a product name means. It is a trivial problem to solve considering it is only a problem due fraud or fraud adjacent activities.
heres the solution: You have your product name, and then any changes to it REQUIRE a new version name. Like how you have Coke and Diet Coke, only formalized such that Coke cannot legally replace the ingredients of regular Coke with the ingredients of Diet coke and still call it Coke. With Software they could at least use version numbers and release modifiers. most of the time its built in already, but it needs to include the full stack. if they change anything that has an interaction with the user it should change the label seen by the user for the product.
You don't buy a hard quantity of an LLM you buy access in general.
A Opus 5.5 is not a Opus 5.5. Its Opus 5.5 of some checkpoint.
mRNA btw. was well known how it worked, it wasn't hidden and some people waited for the non mRNA version because of it. I never seen or read the argument you make and I do not get it.
A vaccine is something which protects you from an infection. What definition do you think changed in humans? That we make the protein? How is that so different from viral vector vaccines?
The consent is a certain level of trust from the society to a group of researchers limited by being humans and having to act in a certain time frame.
Nontheless, I do agree I would like to have a fingerprint and proper changelogs etc. for products I buy including digital and analog. But this has a downside as usual: The smaller you are, the more it costs you to (probably) register, maintain and change.
You know the good thing about Claude etc. right now? It doesn't matter. My invest is not 4 digits and there are that many changes, i'm quite happy to have been upgraded for 'free' from Opus 4.8 to Opus 5 (what do you call this upgrade, which was significant better?)
I agree you buy access in general and they are free to cahnge their offering to what that access is however much they want. there should also be communication regarding what it is you are accessing. the meaning of "Opus 5.5" is based on unwritten trust that can be broken at anytime without consequences. the fact that its a hobby and not much at risk to changes
I also agree that if its too formalized with registrations and such that it could be a negative pressure on small business. documenting what the labels mean could have separate rules and smaller business could operate on trust and face penalty if counterexamples are shown, for example. obv this would be the start of a much longer conversation about implementations but i think it's important
--
with respect to the covid vaccine, i want to be clear that i dont pick this fight lightly, I love science. The rare side effects of the mRNA therapy did actually kill people due to severe myocarditis. I believe it is unjust to change who is at risk without doing everything you can to inform people of exactly what it is they are getting. They treated the mRNA vaccine like we are Gods commanding the body to do whatever we want and not like we never have actually fucking know if something is going to kill someone or not. People can be allergic to things, for example, and its only okay because they themselves agreed.
>A vaccine is something which protects you from an infection
this is the new definition.
the old definition was accepted as: "a preparation of killed microorganisms, living attenuated organisms, or living virulent organisms that is administered to produce or artificially increase immunity to a particular disease"
this aligns with informal understandings and how it was taught to people. you get 'infected' with a safe amount of the actual virus to trigger your body's natural defense mchanisms
the definition was changed in 2021 to mean: "a preparation or immunotherapy that is used to stimulate the bodys immune response against noninfectious substances, agents, or diseases".
informally, and biologically, this is fundamentally different. i admit it's a grey area, in that the word vaccine was used to describe it in research phases.. but it was 50-50 and nearly always included 'mRNA' in the label.
But it was still deliberate misdirection that over-emphasized calling it something common and familiar and omitting whats different about it to push a new thing (new to the public, as it would be almost everyones first exposure to mrna therapy) and not only did they refuse to emphasize the differences and make an effort to educate people but they conflated anyone who asked questions about it to be the same as an "anti-vaxxer" and pushed the narrative such that the new 'vaccine' could not be viewed as its own thing. And then they changed the literal definition of the word vaccine to 'cover the tracks'.
it was backwards. everyone should have been informed, proper consent collected, and i think it would have been better received overall. whether people can articulate the problem as i am or not, they felt it. I felt it, and spent a lot of time and energy developing my understadning of teh situation as to be properly nuanced in my opinion of it and not actually go into conspiracy theory territory.
I still talk to people today who have no idea the covid vaccine is not dead covid virus being injected. They should have been informed that they were receiving something different.
again i never said mRNA didnt work or wasnt effective. but it was still fraud. they wanted to move fast, as if all other medical progress was somehow not operating with lives on the line already and that informed consent was just something we do for fun. Covid was serious and required a serious response with funding to scientific education and research, not lazy manipulation tactics. There should have been 24/7 news about mRNA therapies and their history, booths popping up everywhere to answer peoples questions, etc.
The response was lazy and fraudulent and sets an extremely dangerous precedent of allowing future reactions to scientific education problems to be handled through authority rather than education.
--
If you want to get me started on an actual political issue of the same form to understand why i havent let it go nad only become more emboldened with my opinion since covid, go research how the use of the term 'domestic terrorist' has changed over time. The laziness and use of authority to make things be defined however an authorty wants rather than through actual public consensus is only ramping up.
the danger is that the public is slowly being cut from the conversations completely. when a patient is too uneducated to understand any of the treatment you are giving them, is it still consent?
I really don't know in which country you live, but while we had plenty of discussions in germany if there needs to be a mandate for vaccination, it did not happen.
there was no forced vaccination besides Bundeswehr (our army).
So first off no one was forced, secondly if you told someone that we give you a virus which will give you an immune reaction or a weakend pathogen or whatever, what does a normal person do with this information anyway?
No human in history thought "oh that sounds bad/good lets do/not do that". They trust science and math and thats how it was with covid too.
Besides the fact, that mRNA was discussed up and down in germany at that time. It was brought up in different tv shows, news articles, blogs etc. Every doctor has to explain to you how stuff works if you asked and the internet exists.
So we had a virus, after the SARS outbreak from a few years back were people in canada died, a virus moving very fast and overwhelming hospitals left and right and researchers having a solution which was statistically better to have than covid at that time.
Theh images of italy or New York city went around the globe. A society expects their hospitals to work and be functinal and not overrun by a lot of peple who are dying.
What I do not understand from your position: why do you think adults can't be bothered to inform themselves but need 24/7 news coverige about mRNA? The Wikipedia article alone about this was up-to-date all the time. The people who took the jab of mRNA, were not forced and did not inform themselves and are now com[laining have nothing to complain about.
I got all vaccines, including boosters and i have read up on mRNA etc. i still think it is and was the right decision.
Having a virus in your body which attacks random nerve cells around your whole body is not something i wanted.
I find it hard to see how Chomsky has a leg to stand on.
> I understand how Chomsky might arrive at the conclusion that probabilistic models are unnecessary, from his study of the generation of language.
I think the point turns here for me. We have probabilistic models now that generate unbelievably convincing language. Norvig's following argument on the value of interpreting language statistically becomes secondary.
In order to stay moored in reality, our models must follow from evidence. I think it would take a peculiar obstinance to deny that the latest models show evidence of generating language.
The models require orders of magnitude more data to learn language than children do, and children's task is harder as they have to segment speech themselves.
The comparison isn't so straightforward, since children's language-learning environment is multimodal and extremely rich. They're exposed not just to sequences of words, but also to behaviors, emotions, social interactions, and the physical world.
I think you just explained the situation, even if you didn't realize it.
LLMs process millions of times more text data and use millions of times more compute power to learn a language, mostly because we know how to do that technologically.
Children use much less text data, but also use a lot of other kinds of data, including interacting with other humans, to learn language. We have no idea how that works. And, on top of that, people can interact with pets of various species in the same way as children, but those pets never learn language.
So, something strange is going on that children can do, but that LLMs and puppies can't. Probably, part of it is that human children are much better at learning by copying what adults are doing, and part of it is that humans evolved to learn and use language. The details of what's going on there is what linguistics is about.
Even 10 years of video data is a lot of data, along with sensory and force/acceleration data that every child has as inputs 24x7. And then a few billion years of pre training.
Yes, I believe arguments like GP vastly underestimate the amount of data involved in children's learning. All the data coming into their sensorium, from heard language to physical movement feedback, contributes to training all their skills, from physical movement to language.
A newborn brain is not a blank slate, it comes with a very rich prior (as Chomsky himself would argue), for example the ability to rapidly acquire language, control a body, etc.
An LLM must construct all of its grammatical structure, world models, etc. from scratch during pretraining.
LLM pre-training is more like recapitulating evolution of the mammalian brain than childhood development.
I agree that is something we need to learn a lot more about, but a human child's brain does not appear to be necessary to learn language and I argue the evidence supports that.
The question isn't "are neural networks just like human brains?" It is "can statistical models generate and understand language?"
The answer to the actual question under discussion appears to be a resounding yes. It does not appear that learning language is innate in the way that Chomsky has argued.
It may be that learning language as efficiently as human children will involve some innate property that we have not yet discovered. But that is a perfect example of moving the goal posts. We go from "there is something special about the brain that allows us to learn language, something that cannot be reproduced statistically" to "there is something special about the brain that allows us to learn language efficiently, more than statistical models".
The idea that there is some level of statistical inference involved in language acquisition is not contrary to generative thought, look at Charles Yang's work or Lidz or a fair number of other people. Chomsky just thinks this is uninteresting so he is dismissive of it as a central explanation for language acquisition.
The central question remains whether there are categorical rules built into the brain for grammar (plus "some" statistical inference possibly), or whether the entirety of our grammatical processing is statistical. It is entirely possible that evolution encoded statistical priors into our brains with grammar overall being a fully statistical process. Therefore, the fact that children learn more quickly isn't really hard evidence one way or another.
We could innately have built-in statistical models, then everyone can be happy. Given how well statistics seems to work for language acquisition the brain is probably approximating a statistical formula at some level.
LLMs (at the autoregressive pretraining stage) are trained to predict what an (adult) human who could have said preceding words would have said next. You should add 15-30 years to the human training time.
A common complaint I've seen (and I experience on occasion) is AI agents being a bit silly in their overprotective coding style. They nit pick and catch every single little error. This can be frustrating since a lot of the time they are trying to handle edge cases that are so unlikely to happen, yet to cover them requires significant additional code.
This might be the beginning of the reason we shouldn't be reading the code. It's like in chess engines where the players complain about how weird, alien and unintuitive the moves are. But 100 steps down the line the engine finds some edge case exploit that gives it enough advantage to win.
So we may be crossing a line, right now, where the non-intuitive alien code being generated by agents is seeing edge cases down the line that might allow exploits. And if we "correct" the agent, we may be leaving ourselves open to exploits we don't see and have difficulty understanding.
The answer is not to reduce our understanding. That is the worst possible conclusion we could draw from what we're seeing. The answer is to increase it.
I am saying that we won't be able to write it, in the same way that Magnus Carlson (or whoever your top chess player would be) cannot compete with a chess engine.
I am arguing that, much like the "best" chess move is unintelligible to best chess players, truly secure code will be unintelligible to humans.
It's interesting to see Platonism and virtue ethics combined in such a way to try to make sense of what is going on.
This egoless attitude and the call for a broad humility may find itself in conflict with a supremacist humanism. There are people who will only ever want things that originated from a human. I see this attitude mostly in creative fields like art, music, film and games.
I therefore think we need to have different attitudes towards AI adoption depending on the field. In the case of math, some seem to want to excise any human emotion or physical need and create a pure refined space to explore propositional Truth. But extending this to all human activities doesn't seem to follow.
I guess what I'm saying is, there is probably still a place for human ego, even if that place is not math.
So many people here are opining as if they are correct. I have a large amount of doubt and I like the fact that the world is structured so that people can try things out and see if it works.
If the world was run by the naysayers in this thread, almost nothing would ever get done.
reply