Besides ad hominem and >leave the trillion dollar corpos alone!!Do you have any actual technological arguments against what he said? For example how many times will we hear 'AGI in two weeks'? or conveniently redefining what AGI means every two weeks, why does that happen? Or why is it called 'AI hallucinating' instead of 'blatant lies meant to keep you re-prompting wasting tokens', a SEO tier scam technique old as time.
>>109669745Two more weeks
t. ed zitron, feb 19, 2024https://www.wheresyoured.at/sam-altman-fried/
If AI didn't generate profit they would tell us
>>109669745I don’t have any arguments against what he said because I simply refuse to listen to that grifter
>>109669745I paid $340 a month for this with Fable 5. I am getting very tired of it
>>109669745>For example how many times will we hear 'AGI in two weeks'?Never from reputable people. I hear it often from drooling retards on Twitter who used to be into NFTs, but never from the labs or lab leaders that represent AI labs. The closest we got is Dario saying internships will fall by 50% in half a year which if you compare them to 2022 highs he was partially right, and then some Microsoft CTO or some other position that talked about 50% drop in all jobs in some other number of months which was obviously wrong even back then. But Sammy and Dario and even Elon always said somewhere 2027-2030.
>>109669745Just because people overhype a recent innovation doesn't necessarily mean its vaporwareInternet was hyped in the 90s as the next big thing, then crashed and later actually became the next big thing. AI is on a similar path, it can provide value same as Internet did just not as soon as its touted to be
>>109669807if it ever progresses to actual AGI (doubtful) then there is no guarantee it would be profitable
>ask gpt to check a paper to see if it's possible to implement and commercialise a paper>gpt flags license, tracks down the patents protecting the tech and the author's website highlighting that shit>then offers to construct a version that would likely not infringe on the patentlmao
>>109670149lmao
>>109670185>DarioDidnt he also say "white collar jobs are gone in 6 months" in december last year?
Yo you just gave me the sickest clickbait thumbnail, “A.lie” but y’know done up in a way where the lowercase “l” looks like an uppercase “I” and so it’s like AI when looked at one way but A lie is inherently baked into it damn I’m gonna get SO much negative attention and contribute to the world with this one
>>109670742not really
>>109670149trolled lol
>>109669807if companies used AI and suddenly their output magically got better, they would tell us that as well. for example software kinda sucks ass more and more. there should be billions of bugs fixed to the point of any piece of software should be perfect.
>>109669745>Or why is it called 'AI hallucinating'Because the technology quite literally has no concept of truth or correctness. It generates output tokens that look like they'd match and it learns this from its training set. If you're asking it things that are strongly represented in the training set, then it will likely spit out text that is accurate to the training, but if this isn't the case then the model doesn't have some built-in capability to stop, it will still spit out tokens at you which look plausible but might not be true. That's the hallucination, it's a fundamental flaw of the core design of LLMs.
question, if you ask an LLM to give you a random number, is it truly random?
When I read Xitter I see an army of dumb chuds who swear AGI is already here.We're so cooked.
>>109669745>why is it called 'AI hallucinating' instead of 'blatant lies meant to keep you re-prompting wasting tokens'The latter would imply intentionality, meaning that we had invented AGI.
>>109669807https://isaiprofitable.com/
>>109669745This nigga said AI peaked in 2024 and that the bubble would burst no later than Q2 2026. He claims he doesn’t even use LLM’s so he doesn’t know what he’s talking about. Nvidia’s chips will release this year, costs will halve, models will continue to use reinforcement learning to improve and become cheaper, more synthetic data will be produced, old and bad data will be pruned, and the profits will start rolling in. Anthropic didn’t even plan to be profitable until 2028 but they were profitable in Q2, the quarter that Ed thought the bubble would have burst by now. He is still stuck in 2023 and isn’t adapting at all to what’s happening. He thinks it’s only improving in the ability to “take a test” (benchmarks), which ignores the superhuman math and coding capabilities that aren’t just studying for a test. This guy is either a total idiot or just capitalizing on the anti-sentiment to charge readers for his blog. All he can do is point to large capex and old understanding of how the models world, apparently forgetting that it’s normal for large companies to not profit in early years. He’s even been pivoting to the desperate line that “70% of AI revenue comes from TWO companies— OpenAI and Anthropic”…. Ok? What’s the point? They’re the best companies right now and they’re making lots of money and… that’s a bad thing? This guy doesn’t know what the fuck he’s saying
>>109672393>Nvidia’s chips will release this year, costs will halvethanks to the memory supply crunch the AI companies themselves caused the chips are more expensive than ever>they were profitable in Q2I highly doubt they were profitable with how much datacenter money they've been burning. I suspect they cooked some books or they did some circular financing shit to move the costs to somebody else. (e.g. some big tech company giving them free or discounted time on their servers or something)>which ignores the superhuman math and coding capabilities that aren’t just studying for a testtrue. He also ignored its superhuman ability to rm -rf entire production databases, and its superhuman ability to prove some obscure austistic theorem nobody cares other than some post-PhD math researcher, but it didn't even do that properly, it just gave a couple of counter-examples to some conjectures and reduced some lower bounds a bit.>He’s even been pivoting to the desperate line that “70% of AI revenue comes from TWO companies— OpenAI and Anthropic”…. Ok? What’s the point?The point is that if those two companies collapse, even if one of them collapses, it destroys all demand for chips, datacenters and all that capex big tech companies have been burning for "AI".
>>109670756God's work little brother...
>>109672521> thanks to the memory supply crunch the AI companies themselves caused the chips are more expensive than everI was referring to inference costs, which have been getting hundreds of times cheaper with each year > I highly doubt they were profitable with how much datacenter money they've been burningah, got it. If their numbers look good, they MUST be lying. But Ed never lies about anything or gets anything wrong, nope.> He also ignored its superhuman ability to rm -rf entire production databasesskill issue + this almost never happens + the models won’t be doing this next year, certainly not years from now.> but it didn't even do that properly, it just gave a couple of counter-examples to some conjectures and reduced some lower bounds a bit.wow, you’re actually just misinformed. There are at least 215 AI-discovered proofs that are NOT counterexamples. https://vibemathed.com/?result=proved&contribution=ai-discoveredThe fact is, anyone working with math WILL use LLM’s in the future, especially mathematicians. When it objectively proves that it’s intelligent, your cope is “but that’s just autistic stuff” which is what you’ll say when it masters coding as well. > The point is that if those two companies collapse, even if one of them collapses, it destroys all demand for chips, datacenters and all that capex big tech companies have been burning for "AI".No it doesn’t, dumbass. This assumes that no other companies are buying chips, which is false. Even if OpenAI “collapses”, Microsoft will swoop them up anyway, and at the very least, stop going heavy on training and research, and just make profits on inference. But then we’d still have Meta, Spacex, Google, the Chinese companies, etc. On one hand you say “boo hoo RAM is so expensive because of demand” then you try to argue that it will somehow lose that demand if a single company fails. Get your head out of the sand
AI sycophants are the most annoying people alive
>>109670742He didn’t, but fags like the one in OP pic pretend he does say it and then make ass of themselves trying to compare Mythos/Astra cyberattacks with GPT2, while also making the same arguments that they made when they said they shouldn’t release GPT2 to the public.
>>109670149Why are you paying out of your own pocket for the most overpriced and overrated model on the market?
>>109672633>I was referring to inference costs>at the very least, stop going heavy on training and research, and just make profits on inferenceYes and as we all know inference costs run on thin air, and no training is needed, who cares if a model is perpetually stuck on information from 2025, right?>ah, got it. If their numbers look good, they MUST be lyingThey might not be lying strictly speaking, but I'm highly suspicious of all the asterisks, in pic related for example.>Even if OpenAI “collapses”, Microsoft will swoop them up anywayWith what money? In case you haven't noticed all big tech corporations are taking huge amounts of debt and using up all their free cash flow to build datacenters and currently OpenAI is paying tons of money to Microsoft to inflate Microsoft's cloud computing earnings. If OpenAI collapses and Microsoft absorbs them, how will Microsoft keep their cloud computing earnings up? By paying themselves? And how will they pay for the massive debts they've accumulated?
>>109669745I have no idea who this guy is or what he said.>AGI in two weekThese companies are running on investments. They have to bullshit to get people to throw money at them.https://youtu.be/tZmIK_IqL4Y?t=584>AI hallucinationsAgain bullshitting. It's like talking to someone who is really smart but also doesn't know how to shut the fuck up.
>>109669745True AI is literally a genie that can grant any wish, a solution for every problem, and we cannot allow China to achieve it first.Ten trillion more dollars of investor money, please.
>>109672940> Yes and as we all know inference costs run on thin airinteresting to respond to “inference costs are getting cheaper” with “do you think it’s FREE?!?!”> who cares if a model is perpetually stuck on information from 2025, right?for most purposes it doesn’t really matter no. They can search the web anyway > With what money? Microsoft already owns a portion of OpenAI. If their business model is collapsing then why on earth would they be an expensive buy?? Oh right, they will never actually collapse in the first place because investors know this is the real deal.
>>109669804What the problem. he was correct. There's been no significant improvement since ~4o
>>109670838the technology you are worshipping says you are lying.
>>109673434>said in may not december>in one to 5 years, not 6 monthstime to retire this meatbag, it's constantly hallucinating
>>109673389>not getting betteryou’re right, they’re just paying hundreds of mathematicians millions of dollars to solve all these math problems, yes, that’s it!>durr it’s just maffs!and yet you couldn’t do it. Not even in a million years. AI is smarter than you. How does that make you feel?
>>109669804>>109669745Once you even somewhat understand how LLMs work, you realize that their "intelligence" is a happy coincidence from creating very very complex semantic search. Of course, they don't reason. They don't think ahead. The "reasoning" output from LLMs is so fucking cringe and gay because they're not actually thinking, that's not how they work, even at the frontier level. It's all smoke and mirrors, a parlor trick that yes can be legitimately useful, because it is a very complex and powerful semantic search engine after all, but it is very obviously to anybody with a brain not actual intelligence.
>>109673389In February 2024 we didn't even have Google Veo, much less Sora. Now Sora is dead outdated shit
>>109673665Did you actually type that completely unoriginal post
>>109673665wow, what a miracle that the slot machine stochastic parrot solved the 87-year-old Jacobian conjecture!
>>109673740Luddites are weird bro
>>109673788>wow, what a miracle that the slot machine stochastic parrot solved the 87-year-old Jacobian conjecture!Is that what really happened? Every time I read an article of AI solving something it was a team of researchers who used AI to help them, the AI never autonomously solves anything. You don't call Google AI because researchers use it to help them do stuff.
>>109673832Where have you been? This guy prompted chatGPT to solve the problem by just saying “keep going, you can do it”https://chatgpt.com/share/6a60b2eb-0b64-83ee-9c76-7931ca1de063If you look on this site, there are hundreds of proofs, most of them not done by OpenAI or Anthropic’s research team. At least 200 are AI-discovered, not assisted. Even if they only solve the problems with the help of researchers, it’s clearly a huge improvement in the number of proofs being solved.
>>109673832>>109673884forgot link:https://vibemathed.com/
>>109673884I have no way of verifying if this is real or bullshit. I've used the same models to try and build software and it is fundamentally retarded beyond a certain level, so I am very skeptical of the level of "proofs" the AI is capable of generating.
>>109673896believe it’s all a conspiracy then, I guess. Humans just magically started cranking out hundreds of proofs to make AI companies look good or something
genuinely the funniest thing to come out of recent math results from llms is people who believe that anthropic and openai have a team of math avengers squirrelled away, solving decades old math problems for them lmao
>>109674019it’s the deep state genetically engineered clones. They’re grown up now and they’re doing serious work.
>>109673156Who is paying you to post here?The issue is that this grift has consumed so much capital that it must substantially grow the economy to see a return on investment. LLMs are useful, but they are no where near as useful as previous developments that lead to financial crashes. If you think that LLMs are more impactful than railroads and the internet it only proves that you are a moron. Even if every CRUD app for the rest of time is shat out in a perfect state by an LLM it won’t earn enough to prevent a crash.And to address the final cope: LLMs will never be AGI. AGI may be possible, but we need to see fundamental breakthroughs for it to happen. If we get these breakthroughs LLM investors are even more fucked because their mountains of purpose built GPUs will be worth less than the sand they are made of.
>>109669757more likeFive minutes.
Whoever builds the first robot army wins. What a great time for venture capitalists to be alive.
>>109673896sounds like a skill issue desu
>>109674461Once someone builds a robot army, someone else builds anti-robot weaponry and the race goes on forever. You can only hope to be a few months ahead of the opposition. Look at drones, they have to keep innovating and reinventing shit just to be useable in a war
>>109669745AI is only good for memes, comer stuff, waifus and to aggregate information. Everything else they promised is a scam
>>109673884>solved it by continually trying different methods that already exist in the literatureFTFY
>>109669804> don’t knowAI doesn’t even know that it doesn’t know, so that failed. It just spews out bullshit.At best, it can respond something like a search engine when the word can’t be found
>>109674331The internet took several years before it really became commonly used. It took decades to have the polished Internet services that we have today. But LLM’s are roughly 5 years old and they’re still in the R&D phase, along with advancing chips. The business model was always predicated on long-term improvements. Anthropic didn’t plan to be profitable until 2028. The investors knew what they were getting into. The bubble pop for a few years at least, especially considering, again, chip advancements like the Rubin chip that are yet to be deployed, and various mega data centers that are in construction. We’re just getting started. OpenAI is sitting on an internal model that churned out a batch of 10 novel math proofs a few months ago, and this model will be used to iterate reinforcement learning at a faster pace than before. Will it become AGI? I don’t think so, but it doesn’t need to be AGI to be superhuman at math, coding, and develop an extremely efficient model that servers the average user that just wants to use ChatGPT as a better Google search, or math tutor, or translation service, etc. They are already making billions in revenue, so there’s clearly demand as it is. This will only go up as the capabilities increase, and people become more reliant through habituation and also skill atrophy. Everyone will be hooked on AI, tons of loners use it just to have someone to talk to, some of them even think they’re in a “relationship” with it. It’s not yet clear how much people will be willing to pay, but if the AI companies were actually in trouble, they wouldn’t be handing out the service for free, especially not chatGPT Luna 5.6 which is now free for a whole month. And of course any employee that has to make any sort of document, resume, spreadsheet, graph, etc. (boring shit), then they will certainly depend on AI. Just imagine the current high school students, who no doubt use AI all the time. You think they’ll just give it up? No, they need it
>>109674546that’s what humans do, dumbass. And you think your intelligence is so special compared to an LLM. You couldn’t do any one of those proofs, certainly not in a few hours.
>>109669745llms are integrated into work life and casual internet use whether we like it or not, but yeah theyre not sentient or "agi" as in the idea of an independent lifeform consciousness with autonomy. its not going to "replace every job ever" or cause yudkowsky scifi apocalypse fantasies, but it will streamline many tasks/types of work and has already, and theres undoubtedly a social impact due to the parasocial nature of it that should be monitored/studied
>>109669745You people are coping so hard lol AI is the fucking future and present right here right now. Its already replaced you only matter of time I make music with AI and sexy women
>>109674777>llms are integrated into work life and casual internet use whether we like it or not,wrong, dipshitif the internet vanishes tomorrow the world would be chaotic for like a decadeif LLM vanish tomorrow nobody would give a shit and there would be zero consequencesthis is why i always laugh when retards compare LLM to the internet.no internet = huge global issuesno LLM = oh no we are going back to pre-2022 era where we didnt have slop oh no oh no whatever shall we do
>>109675252this>>109674777>its inevitable get used to it goyNo
>>109674331>>109675252Cope. There is no going back to the pre AI era now that it exists. It saves a huge amount of time. Rejecting AI is a fringe position that is perceived as a sign of socialist views.
LOL at people (maybe disruptor shills) missing the entire point. It does not matter if AI is useful right now, what matters is that the numbers are impossible and the math wins in the end, AI would need to become 1,000X more expensive or 1,000X shittier in order to justify the current investment sunk into AI.
>>109675303>It saves a huge amount of time.thats a huge gap from the >makes you 999x developeras they promised every two weeks and 90billion spent.where is the exponential leap in productivity regarding the end results?
>>109673665Potent reminder that claims of understanding how LLMs work must always be followed by claims of understanding emergent complexity, or else it's worthless.
>>109672393The projected earnings for the big players in this industry are absolutely ridiculous, for them to meet shareholder demands would require a fucking miracle for the next 3 years to justify current valuations. Its not to say the tech isn't valuable, but as anon here previously said, the money bubble might pop even if AI itself is here to stay, much like the dotcom bubble (and yes I know this isn't exactly like the dotcom bubble, but still)
>>109672940In regards to their use of SPVs and the incestous deals that circulate investor capital amongst a tight knit network of companies like this, to think they aren't pumping up their balance sheets for quarterly and annual reports is foolhardy.
Transexuals hate the term AI because it implies actual artificial intelligence , for them who run on the artificiality or feelings equal truth basis of reality if intelligence can be artificial when AI is genuinely potent at some activities then what does that make of the term of artificial women or artificial sex change for they who are troons that can't even pass Tldr only trannies hate ai
its impossibly overvaluedbubble is inevitable and its being used to prop up the dead us economy. the americans are fucked in the ass
>>109675303You are too stupid to understand the situation . At this point so much capital has been committed to LLM buildouts that even if it were to replace all developers it still would be a terrible investment. That’s why Altman and the rest need to pretend they will replace all workers and transform the world. It’s also why all the “elite” investors have to pretend to believe it.They are burning enough capital to build a national rail network to build warehouses filled with GPUs that will be obsolete in 3 years. If you can’t see how stupid tthat is then maybe you are dumber than a chat bot.
>>109674648Feel bad for those students. They will be completely dependent on a machine that will do their thinking for them and will probably be their only friend in the future. Sad times.
>>109669807Goldman has been telling you that for like 3 years at this point
>>1096753521,000x less expensive*
>>109669807Well they did say that housing prices would continue to go up and everything was ok and that was a day before 08 crash.
>>109669745A lIeAIOh ssshit!!!!
We have to show China we are capable
>>109669757To unflatten the curve
ai is literally sentient AGI is already here
>>109674671The argument is whether LLMs is thinking or not, humans are capable of thinking outside the box and we made boats out of sticks and piston engines out of building water. LLMs are bound within the box of human knowledge and it can't go outside it because it can't think on its own. This is just a fundamental fact of LLMs.
>>109676807>buildingboiling*
>>109669745Is AI profitable yet?When is fully autonomous agent coming where I just have to prompt it and it'll read my mind and work on everything that's required?
