he's actually going to win, isn't he
>>109925987Context hat
Sunk cost fallacy
Post the thing next time OP. Best wishes.https://youtu.be/zbKDmkJPVvI
>>109925987As each day passes he’s more and more wrong. https://danluu.com/zitron/He no longer has the cope that inference isn’t profitable. He’s had to concede that the capabilities are getting better but he still has to deny reality to keep his cult following latched on to his anti-AI stance. Now he’s moved onto the cope “there isn’t enough demand for the data centers” which is simply false. Anthropic actually has somewhat of a compute shortage which is why they had to rent from Spacex, and OpenAI had to pause new Astra subscriptions. If they had leftover compute they would use it to train new models anyway. I really don’t know how he can keep the grift going in the coming months. Astra and Opus 5.5 are insane and the internal models are even better. When Vera Rubin chips are fully deployed he’s gonna start pulling his hair out (that is, if his hair were actually real in the first place)
>>109925987Yes.>>109926977>he invested his life savings into AI stockNGMI
>>109927251imagine missing out on 3+ years of gains because this plebbit grifter has been calling for the bubble pop the whole time. This was the guy who confidently said it would happen “no later than Q2 2026”… LMAO
ai corpos are too deep in financial shit, so he may as well be right.value proposition of even the latest cloud models that retards are soifacing about does not match the amount of investment. Simple as.
>>109925987The only reason he hates ai is because they are building datacenters near muh poor blackerino neighborhoods!
>>109926977Yes this. Don't get me wrong: I'd love for ai to collapse (as it puts my livelihood in jeopardy), but denying reality isn't going to solve anything. It's also interesting it's always leftist trannies that are so much anti ai. See >>109927723.I also found this guy on YouTube called pivot to ai or some shit. His name is David Gerard. I go look him up, and lo and behold, he is a wikipedia editor tranny lmao https://en.wikipedia.org/wiki/David_Gerard_%28author%29 You can't make this shit up.
>>109927723You will burn like the data centers you love so much.
>>109927743>David Gerardhttps://www.lesswrong.com/posts/3XNinGkqrHn93dwhY/reliable-sources-the-story-of-david-gerard>[Gerard] was the sort of person drawn immediately to its [the internet's] promise: an Australian sci-fi fan and early career sysadmin, a proudly weird bisexual and polyamorous goth
>>109927784Why are you bringing this wikitranny e-drama here?https://www.youtube.com/watch?v=yWF3XXanddQ
>>109926977Actually a couple of deep AI insiders leaked information about the Collective Call for AI Slowdown recently. Those companies are in the absolute shitters rn.
>>109928271Yeah, nobody wants to slow down.
>>109925987How? Dude has been wrong about everything and keeps doubling down.
>>109925987Bears always win eventually but until then they are constantly wrong.
>>109928265>Why are you bringing this wikitranny e-drama here?Because he gives the same exact arguments as Ed Zitron.
>>109928315Oh for sure they're not slowing down but they realized their (collective) valuations would plummet if they were to IPO one after the other.
the guy who's been wrong about everything for 3 years?yeah i think he'll get it right this time.
>>109926977Came here to post thisI’m less optimistic than you are, but Zitrone being eventually right about a crash doesn’t excuse the fact that he’s been consistently wrong about everything elseZitrone is playing exactly from the creationist/flat earth playbook, which makes it even more hilarious that Reddit views him as a prophet
>>109926977>Inference is profitableProof?
>>109925987we are currently in the mobile carrier data plan, sim-lock, gold digger phase. he's right insofar that openai and anthropic are making losses but not that AI is generally unprofitable. even deepseek are making velvet with their business model and the next stage will be flatrates. as soon there is surplus in infra these infra providers like nvidia and google will sell it for pennies because the calculation will simply be selling it for something or for nothing. thats why google has started to invest in TPUs as well. boutique providers will go under and noone will care.
>>109929029Anon, why in the hell would anyone spend a thin dime on something like open ai when you can have local ai that doesn't spy on you or use your data to train their llms and price you out of the market with prepackaged replacement?
>>109929104because right now local performance is too slow, waiting minutes for each prompt isnt worth it with deepseek and basically free providers like groq and nvidia which you can ochestrate with a $20 codex or claude sub.
>>109928871https://perspectives.nvidia.com/ai-infrastructure/total-cost-of-ownership/task/faq/revenue-per-rack-ai-inference-datacenter-economics/https://bunbunlabs.com/notes/when-inference-has-to-pay-for-itself/https://www.tensoreconomics.com/p/llm-inference-economics-from-firsthttps://mlkan.substack.com/p/how-to-estimate-gross-margins-forhttps://clusterbid.com/blog/ai-inference-margin-analysis-gpu-provider-economicshttps://multigrid.ai/learn/inference-marginshttps://www.aeaweb.org/articles?id=10.1257/jep.20261506https://www.vistaequitypartners.com/insights/inference-economics-enterprise-ai/https://arxiv.org/abs/2506.04645https://www.theinformation.com/articles/16444Ed is literally a conspiracy theorist who never addresses any of these facts and simply says “they say it’s profitable but I don’t believe it”.
>>109929319isaiprofitable.com btw
>>109929325Inference can be profitable while it also being the case that capex is currently larger than the inference profits. This is basic economics. Does it absolutely confuse you that Amazon was unprofitable for almost 10 years? Oh my god, how could they survive! Both OpenAI and Anthropic are ahead of schedule, their investors knew what they were signing up for. Spending large amounts of money in the short term is literally part of the plan. It is the ONLY play in this scenario. Ignoring the long-term profitability by focusing on the present moment is such a nigger-tier understanding of things.>ay bruh if dat shit make money den how come dey be losing money right now yo??
>>109929374>Inference can be profitable while it also being the case that capex is currently larger than the inference profitsInference yes. But the amount of training to get that inference to profit is not, apparently.
ed said gpu's will get deprecrated and thrown out every 2 years4 years later, h100's are worth more / hour than they've ever beened said they just don't have enough power to run all the gpusthey're just buying gas turbines and jet engines and dropping them next to datacenters and it's completely fineed said ai had peaked at gpt 4 and it would never get smartero1 came out the same month he said itthis board was full of schizos who didn't believe that the new datacenters simply didn't exist because they didn't have money despite literal actual pictures of giant new datacenters being out there
>>109929391>who believed* that the new datacenters simply didn't exist
>>109929382If you think that training costs will always be high, then congratulations, you believe LLM capabilities will improve at a similar rate forever, in which case the training cost will be a trivial obstacle just four years from now. You really haven’t thought this through.
>>109926977Casting spells?
>>109929414Yeah man. Adjusting 2 trillion different weights based on at least 200 terabytes of mass human text will get easier. Surely man. Did you think about O(n^2) attention of transformers before writing this or did you just learn about it?
>>109929374 Care to explains why Oracle declared force majeure?It’s not a good sign. Especially when their reasoning for doing so is BS and not for a valid reason such as war or an act of god.
>>109929427Have you considered applying MAMBA to that problem?
>>109929434Did you?
>>109929429because oracle, much like ibm, is a financial engineering company full of accountants
>>109929438It's a love square.
>>109929458
>>109929429You’re talking about a specific data center in New Mexico. It’s just about government regulation which is indeed out of Oracle’s control. But you’re gonna use this example to say ALL data centers are in trouble, right? Lmao
>>109929470ok but let me design the safety protocols
>>109929391The nominal value went up because of government over spending and orange drumpf inflation caused by the war. Not because the actual value or demand increased. Elon was forced to shut down his illegal methane turbines outside of his Memphis data center because of NOS emission and a lack of proper permitting. I hate disingenuous shills more than regular sycophant idiots.
>>109929492reddit spacing, didn't read
>>109929488What do you do when people call you a namefag or tripfag? Genuinely curious
>>109929496might dox them privately
>>109929507What about me? What to do if I unleash my namefag and people keep calling me namefag?
>>109929495Been posting this way since before pleddit even existed or you got your first ‘puter to access the internet on kid. Get rekt just like this AI bubble will. There is no ROI from AI.
Waiting room. Like purgatorie. No, not like the Tories, Purgatory. Like Sheol?
>>109929536fine, i read your reddit post>Not because [...] demand increased.actual delusion. besides, inflation can't account for the sustained demand. and h100s contitnue to be in data centers today and you can go rent them on spot right now. so rapid depreciation theory is bunk.>Elon was forced to shut down his illegal methane turbines outside of his Memphis data center because of NOS emission and a lack of proper permitting.yes and this is good. everyone except the african actually applied for the right permits and waited for approvals before dropping their generators down, so they're not going anywhere.besides, now that colossus 1 has enough power from the grid, he can just move the turbine to one of his other 3 datacenters.congratulations, you're 12 steps behind again and getting clowned on by people who break rules for profit and then pay a tiny fine.
>>109925987Force Majuere
>>109929586>inflation can't account for the sustained pricing*
All the anti-Zitron people are focusing on the wrong things. Afaik they're saying: AGI is coming so financials don't matter at all..
>>109929586>illegal methane turbinesWhy the fuck are you writing like this?If the data center is illegal and the Government in Memphis, Tennessee isn't doing anything that means the Republicans in Tennessee and the Government in Tennessee are composed of impotent people. They are weak and they are pathetic. They need literal strong, competent, Democratic leadership, like in California. That is exactly what they need.
>>109929602This isn’t true though. Inference is already profitable, so there’s incentive to scoop up as much of the market share as possible by offering the best models, which for now requires lots of R&D and training, as the capabilities clearly have not been exhausted. See >>109929319 for inference margins. Note that the only response was “but they’re not profiting RIGHT NOW!”
>>109929374it's about the unit economics, if every new user of a newer model makes more profit, then it works long term, otherwise it's not nigger-tier thinking
>>109929728meaningless foid plebbit comment. Be quiet
>>109926977predictions are often wrong because they fail to account for what the counterparty may do in response to the exact same analysis. in other words, it analyzes the state of the chess board as it stands and assumes no further moves by the opponent.
>>109929833Maybe it's the retarded nut sized brain, jew
>>109929873In other words, Ed only sees one move ahead like a retard and doesn’t even consider the opponent’s move. Yeah that sounds right. But unfortunately this doesn’t save him from the statement that LLM’s peaked in 2024. It’s not as if AI companies said “well shit! He WAS right, but now let’s magically summon a demon to make these models even better so we can prove Ed wrong”
>>109929929no retard. this means he needs to stop making turn by turn predictions because he's only defined the end state of the game (certain loss) but not how the situation may unfold in the meantime (what retarded hat tricks the companies will pull). it's trivial to observe a losing game once it appears.
>>109929625inference is profitable if you ignore the financials of the underpinning infrastructure. what's ironic is that whilst massive coding gains were found in AI these retards are way too overinvested into infra to ever survive the upcoming inflation -> rate hike onslaught. at the end of the day the real goal of AI is to come up with a hyper-competent strategist and killing machine that might come up with a brilliant, innovative, miraculous idea to defeat china, thus forcing a captive global audience and saving the dollar they had abandoned from saving long ago.
>>109929149Throw more hardware at it. It's both cheaper in the long run and, to everyone with an IQ above room temperature, safer to do in house LLM. >>109929625Anyone can profit if the future is magical and everything is juuuuuuust right to ensure it. But there is nothing on earth saying this will be profitable or a boon for mankind. As it stands "AI" is just a cover for rampant fraud and using wallstreet to build massive surveillance and data manipulation weapons.
>>109926812a minute and a half in and i can't keep watching. here is a fact: you can purchase hardware to run an open weight model until the end of time and all of the costs are known. retards are openai and anthropic have dug themselves so far into debt, and they have set a precedence that they need to create a new model that is somehow better than the last by silently quantizing it before each release and they are now forced to spend unnecessarily on training every month to maintain this. the only way they IPO is if:1. they stop spending so much on training and research to keep their consumer facing prices reasonable, as that will no longer need to be baked into the price2. they start charging the actual token rates. this nobody actually knows, but i imagine it's a little higher than what they are charging now since they are likely absurdly in debt and they need to charge a price that not only gives them a return on the model they trained, but also to pay the absurd salaries of everyone at openaiin other words, the ipo will fall flat on its face
>>109929625Two more funding rounds to profitability!
>>109930048> you can purchase hardware to run an open weight model until the end of time and all of the costs are knownOpenAI can make a much cheaper model through economy of scale. GPT-6 Luna is better and cheaper than any open model and this trend will only continue as they previously were not even trying to offer the cheapest models. They can distill their own internal models and release them before the Chinese can steal from the public releases.> they start charging the actual token rates. this nobody actually knows, but i imagine it's a little higher than what they are charging now since they are likely absurdly in debtSo now you’re just making shit up
>>109927251>He didn't invest lmao
>>109930099i can only speculate, but a frontier ai company charging $.10/$.50 on a sonnet equivalent or $2/$10 on an opus equivalent seems absurdly cheap for how much they spend on datacenters/labor/infra/etc.
>>109930217Their goal is not to make profit right now. They just need to survive long enough to keep training new models. They literally let a billion people use it for free. In a few years everyone will be hooked, capabilities will plateau so training costs will freefall, ads will be everywhere unless the user subscribes, and premium customers will be paying thousands of month for frontier intelligence. They’ll be making tens of billions in PROFIT.
There was a time where Thomas Malthus was seen as both an alarmist and a prophet. Time has proven some things precient and some things terribly wrong. While this guy is very influential in his circles, I think he not qualified to make such predictions and even if you take him seriously there is no guarantee he knows anything in the broadest sense. Also, he's a bona fide liar most of the time https://danluu.com/zitron/.
so do we buy anthropic stocks or not
>>109929995stupid investment right now to buy a local llm capable gpu for $1200++ (5070ti) when i can have most models for free and when you dont know what the next two years bring in hardware or model design
>>109930353>buying at IPOkek, are you retarded? IPO purpose is allowing for current shareholders to unload their bags on bigger suckers. you unly buy after 1-2 years from IPO after it bottoms the fuck out
