>>110203699LLMs would shut down if financial capital wasn't backing them anymore. Local models are a joke.
>>110203699If an AI based economic bubble bursts, it would cause AI to be pretty pricey in the short term. Then other tech companies will buy the AI infrastructure for pennies on the dollar and lead to cheaper AI overall
BUBBEL BIG BURST HÖHÖHÖHÖHPLOPPA SLOPPA DI PLOP MAKE THE BUBBELHÖHÖHÖHÖHÖÖ BIIIIG BURSTA
>>110204383Why? The chips and data centers would still exist and electricity isn’t particularly expensive.
>>110204455I mentally skipped to mention that AI bubble popping would cause some of the top AI companies to go bankrupt (which would allow to buy their infra for pennies on the dollar)
>>110204455The valuation of AI companies is predicated on theoretical future profit made with chips that haven't been manufactured yet and data centers that haven't been built yet. If the bubble pops and the promise of future profit goes away, do you think investors will continue to bankroll Anthropic as they bleed billions of dollars a year? What do you think would happen to those chips and data centers that haven't been built yet?Also electricity is incredibly expensive when you're talking about datacenters. Like millions of dollars a month.
>>110204455>electricity isn’t particularly expensivewhat?
>>110203699>thinks ai will still be around after the bubble burstsdo you even hear yourself talking?
>>110203768> jokeWorks on my machine
>>110205478Inference is profitable. If Anthropic only trained their models by updating basic info and not running huge RLHF experiments then they would make billions in profit. The longer your delusion lasts the harder you’ll fall when the truth hits you in the face
>>110203699>he thinks AI will get cheaperlol, if they can't pay for it with speculative capital, they'll just gouge you, assuming they don't fold up completely.This is why they want to get rid of local open-source models; they can't charge you an arm and a leg for tokens if you can just save up, buy a giga-PC, and use that for like 5-10 years. So they want to regulate away any escape, and THEN gouge.
>>1102036992 more weeks
>>110205707> This is why they want to get rid of local open-source modelsNot even necessary. After the benchmarks removed a lot of public data benchmarks the chink models were exposed as benchmaxxers. That’s why no one uses them to solve open math problems or make video games. They’re not even the cheapest option anymore.
>>110205723Now add in censorship (look at Anthropic nixing REA or even Claude's new body image policy) and a massive price hike, and a somewhat-behind PRC open-source model starts looking way better.Just because they're not the right option _now_ doesn't mean they won't be under different and probable future circumstances.
>>110205496If inference is profitable... why are all the ai companies fucking losing money?
>>110206030Anthropic made $16 billion with $3 billion in inference costs. They’re not profiting because they’re spending money on training and free users. They’re not retarded goyim who think “me make money now!” but realize that they can just keep re-investing basically all of their revenue to keep distancing themselves from their competitors, especially everyone except OpenAI who is the only real frontier competitor, but it’s enough to motivate them to stay on top. It was only due to training advances that so many people have been obsessed with Claude over the last month for game design and general programming ability. Now imagine what they’ll have a year from now. If you think AI will hit a plateau then they’ll just stop going so heavy on training and just focus on efficiency, and using their compute deals, serve the models at scale and still have market advantage. Worst case scenario they stop serving free users, reduce their inference margins (which are already 85%..), and they’re still making billions in profits every year due to increased demand and reliance on AI. You’re really a fool if you think the industry will somehow collapse or that everyone will magically get super powerful local models with minimal cost.
Are you people stupid or something? AI won't stop because of an economical bubble, not after what happened in the past 5 years or so of LLM progress. People didn't stop when AI was running on analog computers for fuck sake.Yes it's a bubble, and no that won't stop it.
>>110206221Not who you're replying to, but I >>110205881 (me) >>110205707 (me) didn't say minimal cost, I said a better one compared to inflated prices to pay data center and VC debts and keep information (more) private. I expect local models to stay behind, but how many people rather would use a 2026!Claude level open model in mid-2027 rather than pay 2027 Anthropic's prices? A lot. Enough for Anthropic and OpenAI to lobby against it.I must, however, note that many EAs (like who runs Anthropic) were genuine doomers long before they had any major financial stake in it, and $$KKKam Altmxn's sociopathy/drive/"ends justify the means" behaviors could apply just as much to "ensure meatbags like me survive" as it does to profit. Some of it could be a genuine risk-management effort; but it's conveniently bound up with their profits, surveillance, and censorship efforts.
>>110206270Local models literally ensure this, even if every early AI company which isn't hooked up with a preexisting megacorp (eg Google Gemini) goes bankrupt, and the CPC cracks down on AI research and exports.
>>110203699>cheaperLLM technology is obviously being commodotized and the long-run value of a reading or writing a token will approach 0.see: open weights, DeepSeek, Mistral etc. etc
>>110203699Probably. The little man will suffer regardless, but I'm hoping Americans don't bail out their rich like they always do.Hoping against hope given how cucked the yanks are though.
>>110206299Their inference margins are so good and they train their models so much that they will dominate the market at every price point. Haiku is 10x cheaper than Kimi. The only reason to use open models is for privacy but most people don’t care about that, certainly not businesses that want to use the absolute best AI capabilities and not something 6 months old.
>>110203699What do you mean when AI gets cheaper? The AI labs won't even be able to operate, much less provide their services at the extreme subsidized costs they're doing now once the bubble bursts
>>110203699AI is cheaper now because the companies are undercutting their competition to build a userbase, when the bubble bursts the prices will skyrocket, and people will call it "enshittification and late stage capitalism"
>>110206406see >>110206779, it's cheaper *now* because they want to get people hooked and winnow down competitors, it's like a breathing competition to see who can stay at the bottom of the pool longest.One by one, or all at once, the major frontier companies will tap out, hoping their users (particularly businesses) are hooked and will stay around no matter how shitty the deal gets.
>>110203699Google employees number goes up this quarter
>>110203699>tech company gets money>sells $40 taxi ride for $15>investors cover the other $25>anon expects the taxi rides to cost $1 when the investors go away
https://share.gemini.google/O9yyANdgrJZKhttps://grok.com/share/c2hhcmQtMg_804f6627-4ebc-4473-8723-eb6b6673872b>AI assets go down>Other assets go up>General goods get cheaper>AI gets cheaperSo basically everything gets better and everything the rich guys built we get at a discount? I'm so glad the AI bubble is popping.
>>110206406>Their inference margins are so good and they train their models so much that they will dominate the market at every price point>IPO expected to open at high junk ratingWell, somebody is lying.
>>1102062211. The expenses are not only R&D, it's also hyperscaling, which also has debt obligations and contractual agreements.2. Revenue is padded by circular deals and is tied to the price of tokens.3. Hyperscaling is intended to lower token prices, which directly cannabalizes revenue.Hopefully you can see the problem:You reach a scale where your circle jerk is all correlated and too big, so it implodes.
>>110205496That's not how training works retard
>>110203768>Local models are a joke.I'm sorry, but they code better than you.
>>110206221Insane cult
>>110203699Jesus Christ, the logic in this is so dumb. This is stunning, even for a redditfrog thread.
>>110203699my work swapped off of claude code onto opencode with grok and chatgpt models while they work on their in house AI that is accessible exclusively through slack. I wish I was joking.
>>110204383>if AI economics fail, other companies will buy the assets and do the exact same unprofitable thing againnot likely anon. it’s funny how many retards still don’t understand whats going on. google might literally fucking fold.
