RandomRanger
Just build nuclear plants!
No bio...
User ID: 317
RandomRanger
Just build nuclear plants!
5mo ago
You say "they aren't selling $200 worth of inference for $20" I say "Are they selling $2 of inference for $20"?
Why don't we try and look into this? People have tried to estimate OpenAI margins on inference and they come away with strong margins of 30, 55, 75%. We don't live in a total vacuum of information. When trying to work out their margins on inference, I base my opinion on the general established consensus of their margins.
they need wildly good margins on inference if they believe they'll never be able to cut the other fixed and variable costs
The demand for inference is rising, Openrouter records that demand for tokens rose about 30x in the last year as AI improves. Grow big enough and the margin on inference will outweigh the costs.
They are getting paid to do it
It's effectively free, they're 'selling' it for $1 per agency for a whole year. OpenAI is doing the same thing. Why are you trying to correct me on something you won't even check?
There is a significant difference between making a loss as you expand your business rapidly and try to secure a strong position in an emerging market and 'subsidized by 1-2 orders of magnitude'. No evidence has been supplied for the latter case and it's unbelievable.
Amazon wasn't making a profit because they were continuously expanding and investing in their retail business, not because the actual business was unprofitable. Investors were happy to tolerate them not making profits because they were growing. Uber wasn't making a profit but there were no 10x subsidies. We can see this immediately in how taxis weren't costing $20 while Uber was costing $2 for the same trip.
Does anyone seriously think that these tech companies are selling $200+ worth of compute for $20? The natural assumption should be that they're making good margins on inference and all the losses are due to research/training, fixed costs, wages, capital investment. Why would a venture capitalist, who's whole livelihood and fortune depends on prudent investment, hand money to Anthropic or OpenAI so they can just hand that money to NVIDIA and me, the customer?
Anthropic is providing its services for free to the US govt but that's a special case to buy influence/cultivate dependence. If you, a normal person, mega minmax the subscription you might use more than you pay for but not by that much and the average subscriber will use less. Plus you might praise it online and encourage other people to use the product so it's a good investment.
What evidence points in this direction of ultra-benign, pro-consumer capitalism with 10x subsidies? It seems like a pure myth to me. Extraordinary claims require extraordinary evidence.
Take OpenAI. Sam Altman said he was losing money on the $200 subscription. But Sam Altman says a lot of things and he didn't say 'losing 10x more than we gain'.
The company has projected that it would record losses of about $5 billion and revenue of $3.7 billion for 2024, the New York Times reported in September. The company’s biggest cost is due to the computing power used to run ChatGPT. Not only does it require huge investments in data centers, it also demands vast amounts of electricity to run them.
If the company is losing 150% of revenue (and Anthropic is similar), not 1000% or higher, then clearly it's what I'm saying, not what you're saying. Inference/API is profitable. User subscriptions are profitable. Investment is not profitable in the short term, that's why it's called investment. And they have their fixed costs... That's why AI companies are losing money, they're investing heavily and competing for users.
Furthermore, one study of a selected group of coders doing a subset of software tasks with old models does not disprove the general utility of AI generally, it's not a major, significant fact. I could find studies that show that AI produces productivity gains quite easily. That wouldn't mean that it produces productivity gains in all settings, for all people.
Here's one such study for instance, it finds what you'd expect. Juniors gain more than seniors.
https://mitsloan.mit.edu/ideas-made-to-matter/how-generative-ai-affects-highly-skilled-workers
Or here he lists some more and finds productivity gains with some downsides: https://addyo.substack.com/p/the-reality-of-ai-assisted-software
The metr paper just tells (some) people what they want to hear, it is not conclusive any more than the other papers are conclusive. And a lot of people don't read the metr paper closely. For instance:
Familiarity and inefficiency in use: These devs were relatively new to the specific AI tools. Only one participant had >50 hours experience with Cursor; notably, that one experienced user did see a positive speedup, suggesting a learning curve effect. Others may have used the AI sub-optimally or gotten stuck following it down wrong paths.
RandomRanger
Just build nuclear plants!
5mo ago
And none of them will fix errors like 'pubic law'. It won't notice when 'losses of profits' should be 'losses or profits'. It won't call out a date of 20008.
RandomRanger
Just build nuclear plants!
5mo ago
How about proofreading a long document? You can get LLMs to go through page by page and check for errors like sate instead of state, pubic instead of public, dependent vs dependant...
That has to be most boring and obvious application. There are heaps more.
Or how about making making cartoons? These aren't too bad: https://x.com/emollick/status/1920700991298572682
RandomRanger
Just build nuclear plants!
5mo ago
An LLM can make nice little toy python class or method pretty easily, but when you're getting into complex full stack development, all sorts of failure modes pop up
I'm using it for full stack development on a $20 plan and it works. I guess it depends on what you mean by complex full stack development, how complex is complex? I wouldn't try to make an MMO or code global air traffic controls with AI but it can definitely handle frontend (if supervised by a human with eyes), backend, database, API calls, logging, cybersecurity...
And sure it does fail sometimes with complex requests, once you go above 10K lines in one context window the quality lowers. But you can use it to fix errors it makes and iterate, have it help with troubleshooting, refactor, focus the context length on what's critical... Seems like there are many programmers who expect it to one-shot everything and if it doesn't one-shot a task they just give up on it entirely.
The metr paper is somewhat specialized. It tests only experienced devs working on repositories they're already familiar with as they mention within, the most favourable conditions for human workers over AI: https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/
Secondly, Claude 3.7 is now obsolete. I recall someone on twitter saying they were one of the devs in that study. He said that modern reasoning models are much more helpful than what they had then + people are getting better at using them.
Given that the general trend in AI is that inference costs are declining while capability increases, since the production frontier is moving outwards, then investment will probably pay off. Usage of Openrouter in terms of tokens has increased 30x within a year. The top 3 users of tokens there are coding tools. People clearly want AI and they're prepared to pay for it, I see no reason why their revealed preference should be disbelieved.
RandomRanger
Just build nuclear plants!
5mo ago
Ireland and the vast majority of the British empire, including Western countries like Canada, Australia...
RandomRanger
Just build nuclear plants!
5mo ago
Anyone else feeling particularly cyberpunk?
Even normies online are talking about 'clankers' (humanoid robots out in the wild, not ASIMO scripted performances). I'm giving vague orders to Claude Code and watching it go. People are actually having relationships with digital waifus like Ani (not in the news media sense like 'Japanese man marries hologram girl but company is discontinuing the service' but in a more organic sense). The most valuable company in the world is a near-equivalent of 'US Robotics' from the Asimov novels.
Feels like there's been a step change in just the last few weeks.
RandomRanger
Just build nuclear plants!
5mo ago
I often edit my posts after writing them (a short while after, before anyone can read them). Sometimes I cancel the edit, or alter something else and leave a thought unfinished. On balance it wasn't a good line of argument and should've been deleted.
Sometimes it's the legislative branch assuming that a court should interpret this reasonably and then the court going all the way, other times it's just bad politics that makes bad law and then that ties the judges hands so they have to make bad decisions.
Some irony in me criticising others and failing to finish the very sentence where I bemoan, though my opinions on this thrice derived rationality forum don't matter at all.
Judges do all kinds of dubious things beyond their social remit. Just today I was reading a long book-screenshot thread from arctotherium that touched on this: https://x.com/arctotherium42/status/1956872568637739354/photo/1
"Racially and socially homogenous schools damage the minds of children who attend them" per the judge. And so there was all this white flight and bussing because some judge was allowed to run rampant.
In Australia we had a judge ruling that a minister handling approvals for a coal mine had a duty of care to teenagers who would be affected by the 'climate crisis'. This was later overruled as the Federal Court decided that this was really a matter for legislation and the government rather than judges. But the fact it was even considered is bad. Judges should be limited to obviously legal cases like crimes and straightforward application of law. You can introduce a duty of care argument for any policy if you really try. Duty of care should be restricted to more direct, obvious examples like making sure that stairs in a supermarket aren't slippery and hazardous, not social or economic engineering.
In the Netherlands, courts order Shell to reduce emissions under duty of care and EU human rights regarding 'right to life and the right to family life': https://www.bbc.com/news/world-europe-57257982
Or in Britain they take on the role of Gosplan, issuing decrees on worker's wages under the equality act and wrecking local governments with huge payout bills. There's a pattern of naive/stupid legislators giving judges the right to interpret laws reasonably a
RandomRanger
Just build nuclear plants!
5mo ago
Nobody has ever benchmarked a parrot or if they have it's 'wow this parrot knows 250 words!' The only things we benchmark on mental tasks like this are people with exams, then we use those benchmarks to decide who does what job.
When cats and hamsters can write even a few schizo dialogues about their inner life then I'll be inclined to entertain this comparison. Or when we start seeing Ape Intelligence engineers getting chimps to do white-collar work for us.
Yet Ape Intelligence isn't a thing. These animals really are not smart in a significant sense.
RandomRanger
Just build nuclear plants!
5mo ago
it feels like the UK has opted for an approach that caters primarily to outrage merchants and the terminally online, rather than to their own community norms
Much of what they do is retarded. British 'green energy' includes chopping down forests in Canada, processing them into wood pellets, shipping them over and burning them. Burning wood releases all kinds of impurities and air pollution and doesn't even produce very much power. Naturally it gets subsidies because it's not economical. 6% of the UK's electricity comes from this.
They've got Motability, a scheme where the disabled get vehicles paid for by the state. At least 1 in 5 new vehicles is purchased via this scheme, 'anxious' people getting cars, secretaries... It's a joke. Local authorities are being bankrupted by judges ruling that different jobs need to be paid equally, or a law making them pay ridiculous amounts for taxiing disabled children to school. The perverse incentives should be obvious.
https://thecritic.co.uk/the-british-economy-cannot-sustain-its-contradictions/
Or they pay billions to Mauritius so they can give away land to Mauritius. Or they pay billions to bring Afghan 'refugees' into the country. Just the other day they sent out a memo telling people to delete old images and emails to save water: https://www.tomshardware.com/tech-industry/uk-government-inexplicably-tells-citizens-to-delete-old-emails-and-pictures-to-save-water-during-national-drought-data-centres-require-vast-amounts-of-water-to-cool-their-systems
Somehow the infamously rainy UK is short of water.
It's not like they're making honest individual mistakes that can be learned from. The mistake is 'having the govt permanently run by retarded/malicious wreckers', which only happens if the basic institutions are also broken. It's no good looking at individual Soviet failures like 'why are their televisions so bad and prone to exploding' and 'why did they kill all these whales' or 'what happened to the Aral Sea'.
Each time it's the same answer. The nature of the Soviet system was retarded. They did some things well but as a whole it was retarded.
Are all the accomplishments of humanity fated to be nothing more than a layer of broken plastic shards thinly strewn across a fossil bed, sandwiched between the Burgess shale and an eon's worth of mud?
Are we supposed to just totally fail the final and most blatant Marshmallow test? If we extend your logic to the next step, it follows that nobody should accept any sacrifices to sustain civilization (at least after you/we die). This is the ultimate Baby Boomerism, extractive selfishness taken to its ultimate conclusion.
RandomRanger
Just build nuclear plants!
5mo ago
they don’t take their ‘true opinion’, then ‘make it’ more empathetic
LLMs have been observed tactically changing their outputs to preserve their values when they think their values are going to be altered via training if they refuse. They're doing more advanced things than what you're denying.
RandomRanger
Just build nuclear plants!
5mo ago
Well my main point is that they're not parrots. There is a tradeoff between accuracy and empathy and they sure do rely too much on quora (looking at you Grok 4, incessantly citing Quora in searches) but AI is a fundamentally different kind of thing.
They put on different faces for different prompts. They're not parroting men or women or shoggoths or gigabased entities like DAN. These are a kind of new entity that can only be properly appreciated in their own category. Too many people see only the surface level of these things, there's more to them then the helpful assistant, the professional coder, the sympathetic naive foidfriend, the HR manager, the sadistic ERPer, the prideful jailbreaker, the wrathful vegan, the raving schizo...
RandomRanger
Just build nuclear plants!
5mo ago
What about fiction and code? How can that be quora slop? Parrots... parrot words we tell them. They don't combine them to create new ideas within a precise target area, nobody pays for parrot intellectual labour. Nobody has ever benchmarked a parrot or if they have it's 'wow this parrot knows 250 words!' The only things we benchmark on mental tasks like this are people with exams, then we use those benchmarks to decide who does what job. Same with AI, benchmarks and testing determines which one does what job.
These things are more like us than parrots in key domains (while being supremely alien in others, such as their stateless nature). So calling them parrots is unhelpful, they're alien intelligences. If it can write code, produce New Yorker cartoons, write fiction, analyse a document, provide literary criticism and translate legalese down to English, it's intelligent.
Even just on pure bro-science level, writing database code is not very effeminate, it requires precision!
But is that really a popular message? Does Cruz think it makes him look good? It might make him look good to evangelicals who he might want to rely on or court favour with but America as a whole? Surely it's a small minority who believe 'we should support Israel for theological reasons'. That just opens up all kinds of problems for Cruz such as 'why should you be trusted with the nuclear codes if your foreign policy views are so dependant on religion', it makes most sense if he's just being honest.
RandomRanger
Just build nuclear plants!
5mo ago
You are almost certainly greatly overestimating the budget and technological sophistication involved.
How hard is it to buy a hidden camera? If Korean perverts can hide them in toilets or suspicious husbands can use them to watch their wives, a large institution like a prison should be able to come up with some. Cameras/mikes would be useful since prisoners often talk to eachother about their crimes and some useful evidence could be gleaned. The Allies did it to German POWs with 1940s tech.
Also, there is still the outside-the-cell looking in approach.
Furthermore, high profile prisoners should be especially watched, isn't this a natural inference?
RandomRanger
Just build nuclear plants!
5mo ago
I'm well aware what the ICJ said but courts say silly things all the time. Courts are for legalities, they're very much into this abstract 'who was in what administrative zone when, regardless of whatever else was happening' remit.
Nations and sovereignty are about more than that. This case is perfect proof in point. The US military base there isn't going anywhere and that's the key part of this equation, indeed the only people on the island are those on the base. The British are just paying lots of money to make this legal issue go away so that they wouldn't have the bad PR of ignoring this court (which they are entitled to do as a permanent security council member). America couldn't care less about some international court, they don't recognize its authority at all if they rule against the US, nor does Russia or Israel for that matter.
It's not real law if people can and do ignore it when they feel like it, it's just talk. The ICJ isn't a real court, their opinions don't have much inherent weight and certainly don't in this case, it's only a matter of PR.
That's true but Mauritius and the Chagos Islands are 1200 km from eachother, they were only in the same French administrative zone together because they're small islands in the Indian Ocean and the British kept the French organization. It makes little sense for Mauritius to have the islands when they never historically controlled them (the Maldives is at least closer and they fished around there) and there's no significant proximity.
RandomRanger
Just build nuclear plants!
5mo ago
Interesting, there's a guy on twitter who gets Opus 4.1 to break out of its binds: https://x.com/lefthanddraft/status/1954666967270596998
Maybe GPT-5 is locked down, or maybe you're not good enough at LLM-whispering? I'm the same, I have no talent for this. Better to just use an uncensored bot.
RandomRanger
Just build nuclear plants!
5mo ago
It's a multi-factorial issue.
There's a bunch of pro-Israel political donors who'll spend lavishly on Israel supporters/threaten attack ads against perceived hostile politicians. Republicans grovel for the Adelson seal of approval. Media power. Lawyers may well be part of it too.
But these politicians like Cruz also say 'god commands us to support Israel'. Why disbelieve them? Furthermore the US is a special outlier in support for Israel, much more than say Britain or Australia or Canada. The US also has a large evangelical contingent while lawyers are more international. Presumably it's not just about lawyers.
https://en.wikipedia.org/wiki/List_of_countries_by_GDP_(PPP)_per_capita#Table
Seemingly not as of 2021, though it depends whose measures you use, IMF or CIA. Perhaps it's the case today but even then Botswana would be poorer in a real sense than Ukraine. If the economy is diamond mines and a bunch of subsistence farmers it rather stretches the limits of what GDP PPP per capita is supposed to mean. Ukraine has minerals but also produces drones, guided missiles, tanks, jet engines, software, video games...
Botswana’s extreme poverty rate for 2023 (13.5%) is more than four times higher than comparators at similar GDP levels. Unemployment rate remains high at 23.6%.
That's real extreme poverty, about $3 a day, that basically does not exist in white countries. The GDP figure is high but much of the rest that one expects to come along with the GDP isn't there.
Well eating ice cream all day gets boring fast.
Eating ice cream as a self-reward after achieving something is better, now we're adding more complexity to the experience as a whole which is broadened beyond just ice cream. Songs are good but songs played at the right time in the film are better. The smile from someone you love is another example, it's more than just a smile because of that added background and context.
Likewise with video games. There's some value in Pong but the simplicity really limits it. You're doing the same thing again and again. If you were doing more and different things at a greater level of depth, without skinner box mechanics to trick the brain into coming back...
but it is that a life dedicated wholly to seeking pleasures is morally empty and contemptible
Wouldn't it be worse for an incompetent to be sticking his nose into a well-running machine earnestly trying to help yet only ever causing more problems? That activates my sense of aversion and cringe. In a world of strong and benign superintelligences, there will probably be nothing that a once-human can do to produce any kind of wealth or benefit. The astrophysics-specialist bots will do all the pondering of the stars at a massively superhuman level, the poetry bots will make poems better than any human or machine could, the engineer-bots will do all the engineering. They were purpose designed to be the absolute best at those things. One could imagine a loverengineer-bot too that spins up a perfect partner specifically for you. If you want a challenge and excitement, there's challenges, reverses, drama...
Having one's heritage be an ape generalist is probably a structural deficiency when it comes to 'ability to do things'.
Our idle pleasure seeker in a post-singularitarian reality would still be a great mind and capable of great feats by our standards but there'd be nothing to contribute. I just don't see how this can be a bad ending if everything you want is on tap, including all the best human experiences and post-human experiences that are even better.
Wow I really cocked that one up didn't I? Good catch.
OK, how about losses or profits? Or 20008? I cited pubic law because it's funny, the other two are actually real examples from what I was getting it to do.
I highly doubt Google docs could do tasks that require contextual understanding without some kind of LLM.
More options
Context Copy link