aqouta
Friends:
User ID: 75
aqouta
4hr ago
Are you really surprised why a technology where its proponents and developers are saying that it has a solid chance of killing everyone and/or consigning the median person into permanent unemployment at best, or creating a titanic financial bubble that's going to bring everyone's retirement with it if it doesn't pay off at worst, is extremely unpopular?
The alternative is that we become Europe, a pathetic retirement home telling itself unending stories of its greatness or the populist have us tear ourselves apart on the pathway to becoming so. The chance of it killing us all is coming either way unless we can get yudd style international treaties to pause/slow down(which I do support but am not expecting to happen). And the likely results of trying it and it crashing don't look terribly different to me than the trajectory we're on if we don't take the swing. A bunch of retirement counts but especially silicone valley investor types get hosed and we all get ridiculously cheap cloud compute for a generation.
frankly the median person has really not seen proportional consumer surplus from the trillions invested into post GPT-4 or so LLM development, and in return all they get is dealing with a tsunami of online slop, vibe-coded software that gets worse even as engineers boast about how productive they are, and higher prices for power and RAM.
I mean of course they haven't because those dollars haven't even been meaningfully spent yet because we haven't built the datacenters. The models people are using are the results of more like the tens to low hundreds of billions of dollars and even that takes time to actually build into products people would like. It takes time for actual artisans to learn to use new tools to produce more than just slop.
At least in China they have the message discipline to tell people that AI is going to be used to improve people's quality of life, but all SF can do is jerk off about the AI-induced permanent underclass happening any month now and how dangerous AI is is going to be; it's obvious why the median person hates the AI build-out.
The median person who hates AI thinks a literal bottle of water is obliterated from the universe every time you query chat gpt for anything. Almost all of the concerns raised by the median anti-ai person are retarded propaganda produced by slopulists who, correctly, think of them as cattle that they can lie to with impunity in order to gin enough outrage to propel them into office, keep them supporting their slop rage social media account or because they're genuinely foreign spooks trying to retard our progress.
For the first time in decades, in my entire life, the US seems to actually be excited about real huge scale projects. We lead the world undeniably in a new state of the art, we have energetic and ambitious figures fighting each other in a titanic race to innovate on projects at never before attempted scale. A site the size of Manhattan! that's small beans lets do it in space! It's just such an incredible bummer how cynical and bitter so many seem to be about it. Truly it could all go wrong, scaling could stop it's definitely within the distribution of possible future that tokens become a commodity in such a way that these investments don't pay off. But if you wanted to make America great again, if you wanted to actually build impressive physical things here at home and show the world what we're capable of, then this is our ticket. Were you under the impression regeneration was going to be a sure thing?
I feel like I'm watching people sabotage nuclear energy all over again over bullshit hallucinated fears or disaster or made up sour grapes about how huge amounts of energy from fission wouldn't even be that useful. We made computers talk like people and they've progressed in only a few years from barely coherent chat partners to genuinely useful junior programmers. I think there's a fair chance this whole thing ends up with all of us dead as well as a fair chance that progress peters out but how can you not be excited?
aqouta
1d ago
It just seems like they should have started a lawsuit and it's really suspicious that they didn't just start a lawsuit. They went down a lot of wacky schemes to avoid doing so and I have no earthly idea why the ceo bothered talking to them at all. Seemed like no one involved had any idea what they were supposed to do, probably because even if you dot your is and cross you ts people won't take you seriously if you're some obnoxious youtube personality doing skits interspersed with your legal challenges. The cops seemed to really fuck up at multiple points though.
aqouta
3d ago
Well, it's happened. The pressures of our household have driven a push to use a fire to help with some of the bitchwork of cooking. There is going to be a push to use fire to cook our meat.
The method involves using fire to heat up a vessel that will contain meat. So I figured I'd get this set up in the backyard so I could be familiar with it. Also, for reasons, we aren't going to be allowed to use metal. so last night I dug a pit, put wood in it and got my humble plastic bowl. I tried to cook the simplest meal, some steak and berries.
The first thing that happened as I put the steak in the plastic bowl and placed it directly in the burning fire is that the bowl started to melt almost immediately. This poisoned the meat. No matter how I moved the bowl within the fire the melting only got worse and I burned my hands painfully in the process. Next I tried to throw the berries into the blaze and was rewarded only with foul smelling smoke.
I know a lot of people reading this are AI evangelist. Where did I go wrong? What the fuck do people see in this shit?
You used tools no one would ever have advised you to use, in a way no one would ever advise you to use them in. There isn't like one or two things that you could tweak to fix this performance. open up a free chat gpt chat and just ask it what you should do to to get started.
aqouta
5d ago
You're making it sound like it's some kind of cynical vote. You risk your tokens if you oppose consensus and anyone can buy the tokens. It's not like a shadowy council decides.
aqouta
6d ago
This ones feel pretty blatant as a resolution for certain polymarket insiders who control how markets resolve via crypto vote, though.
Does polymarket no longer resolve markets through their secondary market mechanism?
aqouta
6d ago
You understand that the title of that episode is referencing the hoax right? It means exposed her as a hack. Didn't dox her, didn't chase her off an platform, didn't try to silence her in any way. Merely exposed her as a hack
aqouta
6d ago
Is the implication here that hoaxing someone in close temporal proximity in time to when they are doxxed makes you also a doxxer?
aqouta
7d ago
You can't be "anti-cancel culture" while also "unmasking" your enemies and urging your followers to to go after them.
He didn't do that, it didn't happen. Taylor Lorenz is the one who doxed LibOfTikTok. If you were just some guy on the internet then this misunderstanding is reasonable but you've been holding this absurd grudge since it happened and surely you should know this.
aqouta
8d ago
In my opinion @TracingWoodgrains has a lot chutzpah complaining about a lack of civility/norms given the active roll that he and his friends at Blocked and Reported played in dismantling those norms. TWG spent a decade knocking down trees to get at the devil and now that the wind is blowing he finds himself with nowhere to hide. BarnabyCajones, LibsOfTikTok, DataRepublican, and all the other right-coded online commentators he tried to shame, dox, or sic a Twitter-mob on over the years send their regards.
What are you talking about dude? That's like the exact opposite of the block and reported/trace/rdrama ethos. Trace hoaxed LibsOfTikTok to show that she was a credulous hack, not to dox or chase her off the internet. You're perfectly happy to see this happen when it's going after lefty institutions and frankly those are farm or often the targets, but you have this absurd and burning hatred the one time it's mildly done against your own hack pundit.
aqouta
10d ago
He has posts on this site where he doesn't do that too tho. I may just be annoyed because he blocked my on twitter after accusing me of being a Maga guy after his flame out here.
aqouta
14d ago
I would expect a first rate scholar to be fluent in the languages in question and be able to actually dog walk some wikipedia warrior on the topic without resorting to petty name calling and shrill moralizing. If that debate performance is representative of what first rate scholars debating looks like then it is a condemnation of the whole concept of scholarship. Genuinely embarrassing. His arguments were twitter fight tier. His debate partner at least handled himself better.
aqouta
14d ago
If you thought finkleestein looked good in that debate then I don't know what to tell you. "Debate is about seriously handling ideas, here see how epic it is that a good debator played name calling games". I don't understand how your mind works.
Destiny is a fine enough debater. at least the two or three I've seen him in, most front of mid was the Israel debate on Fridman. He's not a one in a billion talent but compared to most pundants he handles himself well. To be a good debater you do need to actually handle yourself well in a real time discussion. If there was a glut of people in the rhetoric or philosophy department that could handle themselves well in this kind of debate they'd be able to make a splash in the debate scene, which is fairly open.
aqouta
17d ago
It's a bit messy to make sure you're comparing apples to apples. Here's a breakdown on how deepseek was getting 500%+ margin on inference around the "deepseek moment". Now that comes with a number of caviots, I'd probably hedge that down to more like a 300% margin for deepseek in practice. And they later cut the token cost something like 75% on that model but also reported cost reductions. Gpt O1 was a similar era model(December 2024 VS deepseek Jan 2025) and openai was charging something like 15-30x per token that deepseek was. The model was a similar size but superior in some ways so anyone's guess how much more it cost to serve. That might be the closest apples to apples comparison. I'm pretty confident on a 400% inference margin as a conservative estimate. Inference seems extremely profitable, just that training is also extremely expensive and you need to constantly do it to compete in the inference market.
aqouta
17d ago
you missed the subtext in my writing and so your reaction is quite biased and relies on incorrect interpretation. The others who reacted like you in this thread made the same mistake.
What on earth is the point of publicly registering predictions when you can claim half of them are a joke? In one year when all the 2030 predictions have come to pass there's nothing stopping you from just saying the whole thing was a joke, what a waste of time.
aqouta
17d ago
Surely if you think AI is capping out then you should expect ASICS to be the play. QCOM and the like. I have a hedge in some of those in case scaling doesn't continue as I expect.
aqouta
17d ago
Yeah, people are confused because of the big capex expenses but you can compare what the major labs charge for tokens with what the open source models that anyone can run charge for tokens and notice that the labs have to be taking like a 400%+ margin on inference.
aqouta
18d ago
Your stuff through 2028 is already happening. People are already trying to do basically canned drop in tech workers at every software shop I'm aware of, you're basically just describing Claude code/cursor. Those start ups already exist. Models are already making big progress on erdos problems.
They begin to think about targeting non-coding white collar work like finance and spreadsheet
This has already happened. I'm baffled that you don't know that the models can now handle spreadsheets. They do so pretty well, especially after Opus 4.7.
the models are not getting much better at JavaScript, having used up all of the JavaScript data in the entire world
This is a misunderstanding of how models improve. It's not a matter of finding more undiscovered java script code to ingest, much of it is now post training self play and should continue to improve as general model scale increases. Of course it's already perfectly capable of writing good javascript and has been for several models, the limitations are mostly in reasoning about larger chunks of the code context.
More attention is turned to local models as a result, but these are hard to run on normal hardware and the best is Sonnet 4.6 level at this point and requires $10,000 worth of GPU machinery.
It's too bad Ilforte left because he'd eviscerate this. I tend to be less optimistic on the Chinese models than some but both Deepseek and Kimi have offerings that are comparable to sonnet 4.6 if you trust the benchmarks, I don't but fully expect them to have a sonnet 4.6 level model by end of 2026 and likely an opus 4.6 model by then. And you can run these models on rented hardware for pretty cheap. Although they'd be hard to run locally for a lot of complicated reasons that have to do with it being much more efficient to batch queries than run them individually. In any case though the weights are public and anyone can set up an api to sell tokens at affordable rates.
I'm skeptical of your ability to predict the future as you seem incapable of predicting the past.
aqouta
18d ago
This is a straw man. If anyone critiques the present United States/Western economic system, he must believe in some terrible 19th century egalitarian philosophy book.
You have the same basic problem of the communists, you think you can decide better than market actors what labor is worth paying for. It's an incredible hubris. The employer wants to pay senior staff well, the senior staff wants to do work but you seethe separately at the arrangement.
The problem with this is that basically nobody who makes this argument is willing to let the fentanyl market come to full fruition. Or many other markets.
There's a lot of space between anarchism and not believing that you should get to decide who's labor is valuable and who's isn't for other people.
What are you talking about? In reality I am only allowed, with a ton of friction, to create a tax paying company that panders to the democratic market. If I create a business that actually crushes other institutions and takes, that's called a criminal enterprise in the vast majority of cases, and the commies come in and violently destroy you. We do not live under anarcho-capitalism.
Are you talking exclusively about state run businesses?
aqouta
18d ago
Sure, if you want to quibble over how far of an advancement mythos was then I think there are a variety of reasonable opinions. But this is very much part of chicken's "Ai bulls in shambles" series of posts and I can't help but point out that the people he's calling out as being in shambles here were basically right in previous iterations if you take the supposed doomsayer's opinion at face value.
Yeah, we should get on those bilateral international treaties that Yudd wants. Until then the alternative is unilateral ceasing where we lose all the upside and delay the problem by optimistically 6 months to two years. I don't really see that as a reasonable choice, especially because approximately zero of the people you're talking about are taking x-risk seriously anyways. This subject is more about me hedging my excitement and has nothing to do with what the opponents actually care about.
They are being built, just have not yet finished being built. and no one besides investors are being asked to give them anything. If people don't want to be personally invested in these projects then I have no real beef with those people. I haven't decided exactly what my own financial position will be in relationship to the IPOs.
You're conflating a number of different people and positions. none of the labs are pushing the "permanent underclass" line. That's an expressed anxiety of some people on social media who feel something big is coming and they don't know how to prepare for it. It's like when people get indignant at "tech bros" for the learn to code meme. It wasn't tech bros that proposed miners should learn to code, it was random journalists. Anthropic is currently in the process of getting blackballed by the department of war for taking a stand against their models being used for domestic surveillance and autonomous weapons. The labs and their staff have been unusually honest and forthright with their concerns about how we need to work out as a society what we're going to do if a singularity event happens. For the most part people like op have accused them of fear mongering to hype their product. No doubt if they downplayed the risks they'd be accused of lying about the potential downsides.
I don't know what to do here. Really. I have no idea what I'm supposed to do about people who make terrible and ignorant arguments as soldiers against fantom interlocutors. That's one of the most frustrating elements of this whole topic. There are civilizationally important conversations to be had on the topic. But all I'm presented with are people who somehow simultaneously believe that the whole industry is a scam that doesn't even work while also being an imminent threat to all jobs which it will somehow be able to remove without actually providing any value. And this incoherent position is expressed in the form of on their face ridiculous empirical claims. It's exhausting.
More than anything what I feel about these people is that they just want to be mad, half of them have been whining for years about American decline and incapacity. This is a field where we are excelling, if America is going to right itself and bring about a new age of American prosperity this is it, this is the chance. Should we squander it? Certainly one should be cautious and prudent but that isn't what these people are. They have already written us off and want things to get worse to validate their black pilled priors. They'd rather this all fall apart so they can smugly say "I told you so" in the ruins. I fear some might even, if they could, sabotage us to this end. I want to wake them up.
More options
Context Copy link