Proof that LLMs have plateaued and that there is no AGI coming.Training any bigger models with current methods yields diminishing returns.It's over.
i mean it's obvious, they fed the whole internet to the markov chain, what else is there to do? at this point you'd have to start making up synthetic data
>They have plateau'd>*shows pic proving they didn't*
>>101536379yes, we've been telling you that since a while nowwould you consider shutting the fuck up already?
>>101536407hopefully they can force chinese artists to churn out more work
>>101536379Yeah, GPT-4 is a hard limit.That's the reason why OpenAI isn't releasing new models.
>>101536379Nope. It just means the ceiling needs to be raised. Add in another sets of benchmarks to be met.
>>101536379Sorbet aka Sonnet 3.5 broken through the barrier like butter, L3 purposefully left it out.Wait for Opus 3.5
>>101536632>better models coming>just two more weeks, trust the plan
>>101536379OpenAI was bleeding so much with GPT-4 that they purposefully cucked it to a lower version with shitty benchmarks, and are selling it like it's revolutionary. We're almost at the limit already. It's high time they introduce a revolutionary architecture if they want to continue the exponential growth and not fall into a AI winter, where it'll get significantly cheaper but it'll not get significantly better.And holy fuck NVDA is getting so rich from this it's insane.
>>101536379Good
>pattern recognition engine doesn't become sentient just by force-feeding it moar patterns to recognizepikachuface.jpg
the most outrageous scam in history?AI or the Alaska purchase?
>>101537416>Alaska purchaseRussia couldn't administer the territory because of how far away it was and was worried about Britain just invading and taking it. America was happy to get more land. How was it a scam?
>>101537438so it wasn't a scam but a shakedownimportant distinction
Generative AI is just a mathematical behemoth that calculates the probability of something being true. If you ask it to generate something it just guesses. It's not "intelligence" nor will it ever reach AGI. We're not even a fraction complete in achieving AGI. Sam Altman is the biggest conman in tech ever
>>101536709>It's high time they introduce a revolutionary architecture if they want to continue the exponential growth"They" don't want to do that, because the revolutionary architectures that work (yes, they exist) don't use those expensive nvidia chips. Better you get cucked out of real AI for years than "them" not make their money back.
>>101537479You're not getting your territory back, go be a seething pidor elsewhere
It's not coming, but it may be vcooming.
>>101536379what happens when it reaches 100, singularity?
>>101536467>Good morning sirs, do the needful and use a 10x larger model for 1% better performance