>95% of AI projects are now failing to produce income>80% of AI agents are failing basic taskIs AI a massive grift?
>>107059790>>95% of AI projects are now failing to produce incomeare you SURE its as high as 5% now anon?
>>107059790>citation neededmarket is doing great.
>>107059790AI video is ruining our Information Age while costing us trillions in taxpayer money and retirement plan investments
>A - Always>B - Be>S - Shovel-Selling
>>107059790>Is AI a massive grift?You're just now figuring this out?
>>107059790>80% of AI agents are failing basic taskThis is the one I don't get.How the fuck do models get worse?It is visibly so, but I just don't get it.
>>107059790>95% of Actually Indian projects are now failing to produce income>80% of Actually Indian agents are failing basic taskso it seems, so it seems.
>>107059970With the amount of AI slop published on the internet, the models are being trained on data they themselves created. Going in a downward spiral of qulity and reasoning
>>107059790Yes. It always was. There was no conceivable way it was ever going to make back a billion dollars, let alone multiple.Pic rel is the root of the problem. This is the attitude everyone in tech has. When you hire someone based on how well they can scam and grift people that's all you're ever going to get. They're never going to make anything because if they ever reach a point where they're doing actual work they'll stop and go: "Wait. I screwed up somewhere along the way." And then backtrack until they're getting money for selling big lies.
>>107059970Same way search engines and auto-correct get worse. It very likely reached it's apex long ago, but in their mind "line must GO UP!" so they force them to screw with it hoping they'll speed run a century defining breakthrough every few weeks, but instead they just start breaking it. Now they've reached the point where they think if it's not a coding problem it must be a processing problem and are demanding access to all of our fresh water (yes, all of it) so they can make a 10 second clip of the CEO fighting Darth Vader in ESB in the hopes someone else will cash them out and then it will be THEIR problem. This is the only thing they've done for almost 40 years now. In fact, the "tech industry" wouldn't even exist if it wasn't entirely subsidized by the Midwest through taxes that turn into government grants.
>>107059970I think all current AI problems from hallucinations to alignment to agents to just straight up lying to you come from roleplaying and people talking about AI. People online say chatbots are retarded and constantly hallucinate, that data goes into pre-training and then that LLM is given system prompt that says “be a good chatbot AI and answer questions” and then suddenly that Chatbot starts roleplaying and interprets the “AI chatbots are retarded” as instructions the same way it takes Luke Skywalker is son of Darth Vader as instruction.
>>107059790>I-its a fad I swear!Cope. You arent required anymore luddite.
>>107059790Has any ai project made a single dollar of profit by now?
>>107060261Some scam businesses and hardwere if you count hardwere for AI as AI profit
>>107060261
>>107059970There's no new non-synthetic data to train them with. We've already thrown pretty much the entirety of human knowledge at them.Studies show that model quality degrades quickly when you train a model with synthetic data, train another model with synthetic data made by the previous model, train another model with synthetic data made by the previous model, etc.They keep adding safety shit to the models and the models get dumber and dumber as they add more and more safety shit.
>>107060292bla bla bla who cares it already works. My computer is shit and I can generate french maid AOC with tits out with zero effort, I have created two computer programs that I use almost everyday with free chatgpt, it's 100 percent real keep repeating reddit buzzwords, doesn't negate how amazing it is
>>107060321The image generation is actually still getting better.LLMs (text generation) are plateauing or getting worse.
>>107060326How is image generation getting better when more and more images are AI generated? Shouldn’t the image generation and video generation be already fucked?
>>107060355I imagine it may be because there's still plenty of room to scale them up size-wise because, up to this point, the image models have been tiny compared to the LLMs.Like, you can still run the best anime genning local models on a 12 GB card.They've already hit scaling limits with LLMs where making the models bigger doesn't necessarily make them better.
>>107060261ai derived projects? some, yes. But obviously that becomes highly obfuscated fast (on purpose) by companies forcing it into their existing products and then saying "SEE EVERYONE USES IT!"Actual proper AI/LLM development and research? No.
>>107059970https://andonlabs.com/evals/butter-benchRead the white paper. The model has an autistic melt down in one of them and just gets stuck in a schizophrenia loop.I'm retarded but this is just some of what I've read as a fault point:Context rot,Bad prompts,Bad context engineering as an extension of bad prompts,not using a more distilled/tailored model for a specific task,using LLM when another form of AI would excel,no technological paradigm shift, so answer is "moar compute",LLMs don't have any predictable output because of how weights affect each other,Non-local models getting "tuned" and they're black boxes you have no control over.Researchers don't actually FULLY understand how LLMs are operating when they reason, which is funny. I read another abstract that I can't find anymore where Other cool papers:https://arxiv.org/abs/2503.08908https://arxiv.org/abs/2503.22395https://arxiv.org/html/2507.13169v1
>>107059970Remember the early super competent models all the big AI peddlers were letting you use for free? Yeah these cost a shitload of money to run and they were operating at a loss and cruising on investor bucks. Now free users get the bottom tier models, and paying users get the medium tier models. The top tier models are only used for benchmarks and ad campaigns.They're still operating at a loss, mind you, just not as big a loss
>>107060020the soviet union invented spaceflight with imported foreignersimported foreigners put usa on the moonno i will not name who those imported foreigners were that would be antisemitic
>>107060375Well, image generators have only improved in character consistency and image editing, and the star of that is Nano Banana, which is tiny, so I doubt that. Maybe images don’t improve after all, because I have not seen any qualitative change in the pre-training versions of them for the past 2 years.
>>107059790>Is AI a massive grift?always has been
>>107060006But can't they, like, not train the model anymore? Why are they poisoning their own models and releasing it?
>>107059790none of them are profitable its just an insane arms race because everyone is afraid to fall behind
>>107059790what is the market saying? are they firing humans and replacing them with AI or not?
The modern stockmarket is a fractional reserve-ponzi scheme chimera.
