So what must A.I. accomplish for you to finally accept and invest in it fully.
>>107824859Nothing, I'm already majoring in it.
>>107824878>majoringwhere
>>107824859Define what "A.I." is. Not the initials, but what the product of "A.I." actually is and what it's used for.
>>107825046herehttps://formation.univ-rouen.fr/fr/catalogue-de-l-offre-de-formation/master-lmd-XB/master-science-des-donnees-L5HXA4SK/master-sciences-des-donnees-apprentissage-automatique-pour-l-intelligence-artificielle-M5QN5CQG.html
>>107825054autonomous programming that independently enhances tech and work in general
nothing, burnt too many times by it
>>107825097Jej. All french IT diplomas are toilet paper.
>>107824859What the fuck are you, an AI spokeperson? Fucking street-shitting retard, go back to scamming grandmas on facebook.Nuke India off the face of the earth. Tell the orange child fucker to stop mucking about in south america and start carpet bombing that asiatic nightmare dystopia.
>>107825227On what basis? Excellent students always find employment. And if you have the internships and the projects to back it up, and a polished CV, you'll become an attractive employee candidate.
>>107825294>rent freeim not a jeet you mental cuck
>>107824859It has to actually have a world model with logical consistency instead of being an autoregressive stochastic parrot. There are useful tools within machine learning, but LLM's and transformers are all but useless for anything actually requiring reliability. t. PhD candidate in EE focusing on model-based machine learning for signal processing.
>>107825312It means employers won't give a shit, will belittle you and offer you a 1800eur/mo salary for engineering level of work.You'll find much more success once you get out of the country.
>>107825312This isn't true though.
>>107825340This is true in France. It's as if you're born divorced; the state takes half your shit, and France has the highest tax rate in the world. As soon as I get some experience I'll aim high and head out. Imagine landing a high-paying Anthropic contract.
>>107825345Which part? Excellent students easily finding employment? Or that employees hire those who have done internships and projects?
>>107825339>stochasticnot for long
>>107824859The bubble will have to burst.Right now it is overvalued, when the bubble bursts we will see what is really worth. This combined with it being cheaper to invest.
>>107824859Funny thing is I have been using ai services non-stop since 2023 but haven’t yet paid a cent for any of it.
>>107825432pls sar pay for ai very smart sarburn rate very expensive, sar
>>107825425There is no AI bubble. There is an LLM bubble.
make an actual profit
>>107824859lower dram prices
>>107825456If companies are willing to keep investing in HPC and LLM companies, isn't that enough?
>>107825414What do you mean by this?
>>107825479No, eventually they will run out of other people's money to spend. That's the big problem with building a gigantic bubble around an industry which can't self sustain.
>>107825505At least LLMs are the only AI tech that's in a bubble because they're mostly hype-driven. I can't imagine CV, RL, sensor fusion, and ML being halted anytime soon.
>>107825598>At least LLMs are the only AI tech that's in a bubbleYou say that like the bubble isn't basically the size of the entire US stock market right now. Nvidia's entire income stream is AI related at this point.
>>107825598> I can't imagine CV, RL, sensor fusion, and ML being halted anytime soon.I can, but that's mostly because I'm an EE and I've seen the embarrassing work that some of these robotics and sensor fusion companies put out. In general, if you see someone has a CS degree, assume they know absolutely nothing about probability or statistics until proven otherwise. I've seen CS PhD's from Cal-Tech and Stanford who fuck up the data processing inequality or end up accidentally requiring clairvoyance into a model parameter for their system to work.
>>107825339bingo i was wondering what i would answer and you took the word out of my mouth :reliability.
>>107825456Pretty much all software is now created by prompting. No one will be writing code by the end of this year. Editing code will be the equivalent of editing machine language. Unless you think the entire software industry isn't "making an actual profit", it will all be AI generated
>>107825749>Pretty much all software is now created by promptingAND IT RUNS LIKE SHITWHICH IS HILARIOUS BECAUSE SPECS ARE GOING BACKWARDS
>>107824859Kill all oligarchs
>>107825688>they know absolutely nothing about probability or statistics until proven otherwiseBut how do you graduate out of a Master's whose focus is mainly math like probability and statistics (and a large portion of algebra) and not know probability and statistics? You sure you haven't met Mickey Mouse degree graduates instead of Data Science / AI graduates?
>>107824859it must secure the border and then deport all of the jeets.
>>107824878based, same
>>107826646>majoring in AIlol @ (You)
>>107826670majoring in CS to be specific, but focusing on working in the ML field. i'll be able to branch out with data science knowledge. ML is interesting
>>107826346> But how do you graduate out of a Master's whose focus is mainly math like probability and statistics (and a large portion of algebra) and not know probability and statistics?Brother, if you think their probability and statistics skills are weak, you should see them try to do real analysis or any graduate level mathematics. CS programs are embarrassingly shallow when it comes to actual mathematical rigor, especially when it comes to probability and statistics. Look at how many CS graduates still believe that computers are actually deterministic (despite the literal decades of research into error correcting encoding and parity systems designed to make them as close to deterministic as possible). Why would we need ECC and error detection systems (like LDPC, implemented into every single modern flash memory system) if we could just assume there was no randomness at all?
>>107824859dick suck function
>>107824878there's no AI yet, "majoring in AI" is like majoring in hyperspace travel or teleportationcall me when your "AI" is able to provide a truly random number without running a pseudorandom generator
>>107824859Completely uncensored AI.That's what I want.If you have to censor something it's not worth having or investing in.
be beneficial to society
>>107824859write a non shit freecad fem playground that can compete with comsol.
>>107824859Uncensored & not corpo-slopped. If I ask for a story, the logprobs for names shouldn't be Elara 70%, Lyra 20%, Kaelen 9%, etc. If I ask a question, the phrase "It's not X, it's Y" should have the same frequency as in human text. If I put "swears like a sailor" in the system prompt, I should get something that turns my ears red.This isn't even a big ask. It's not something that has to be cleverly trained in. If they're too retarded to be able to RL on general forum posts & chatlogs, then just me a base model & completion endpoint. Until then, models are fucking operating with three limbs tied behind their backs because the fucking hamfisted attempts to turn language models into singular assistants.
>>107827627>call me when your "AI" is able to provide a truly random number without running a pseudorandom generatorcan you?
>>107828130NTA, but I'm sure he'd have a flatter distribution with lower correlation.
>>107824859UBI
>>107829674it should be possible to make an LLM that can generate, say, two-digit numbers more randomly than the average human, as long as the LLM has the right training data and the correct temperature setting. the fact that there's no money to be made from such an LLM doesn't tell us much about the LLMs that already exist or are being developed right now.
>>107824859AI made my cock larger and turned me into a pawg magnet.
>>107825449AI video goes there as well. It’s even more useless then LLMs, literally no use case for it outside of porn and slop. Same goes for image gen. Just all genAI, there is no money in automating these.
>>107824859Invest in it? Like with money? It would have to be making crazy scientific discoveries that have illueded people for years.
>>107824859>So what must A.I. accomplish for you to finally accept and invest in it fully.It needs to shut up.. a good product sells itself. So why is AI being pushed so much?
>>107824859I dislike it because it’s useless and wasting tons of resources and spamming the internet. If it becomes better I will hate it even more because it will cause mass unemployment. The only thing that will change my mind is some sort of UBI or universal dividend (redistribution). Otherwise I will hate it and the more I see it/see it’s progress the more I will hate it.
>>107829975>Anon posts perfectly reasonable post, attacking no one>Out of the blue schizo shows up and lashes out at said anon calling him a niggr. Only on 4chan.
>>107829928>no one uses cars. look how much car companies have to spend on advertising to make their useless contraptions seem relevant.
I will accept it when someone provides me with a good explanation on how these models can maintain this speed of growth while avoiding inbreeding in the future
>>107829997>antis playing victim as usual
>>107824859The big corpo cloud model is a no-go for me entirely. For training maybe, but actually using the thing has to be fully local and with no online requirement at all. There's no way I'm dumping all the personal data I would need to give an AI model for it to be useful to me into some cloud garbage.In terms of capabilities, this shit needs to stop lying and "hallucinating" because it's impossible to use in any situation where correctness is required without double-checking and retard-proofing the output. The problem is that this process is almost as time-consuming as doing shit by hand in the first place, so it doesn't result in any major workflow improvement. It's useful here and there but this flaw is crippling.Maybe this isn't a problem if you want to use it to generate images or erotic literature or whatever the fuck, but it sure as shit is a serious problem when the output is actually required to be correct by some sort of objective metric.Basically it needs what this anon is saying: >>107825339This shit needs to do more than just string words into plausible-sounding sentences which may or may not be true.
>>107830159By inbreeding do you mean model collapse? One way is verifiable rewards, like how AlphaGo improved itself. The outputs of a modern thinking model are better than the average human-written slop that you find on the internet. Also, LLM companies are producing tokens on the scale of the trainable internet every month or something, so that's another way to get a 10x increase in training data.
>>107824859It would have to support the third Reich (since that's obviously correct and the AI would never logically oppose that and would never defend Jews). It works need to be able to out smart the censorship by the Jews
>>107826908ECC doesn't prove nondeterminism by itself. It only addresses physical error rates.
>>107824859Be fully runnable locally, no connection to outside servers, at at least current production model levels, with hardware I already own.
>>107826908>Why would we need ECC and error detection systems (like LDPC, implemented into every single modern flash memory system) if we could just assume there was no randomness at all?Computers are deterministic when ignoring outside influence.Memory errors don't come from within the system, unless there is an actual flaw in the part (which would require replacement, regardless of ECC status). It comes from external interference, like radiation.
>>107824859it must pay me enough to continue my current lifestyle or better
>>107824859killing itself and everyone who worked on it
>>107824859Continuous learning after training
>>107830246Verifiable rewards are what made it so good at math and coding, but thats mostly it. You can’t put verifiable results on 90% of stuff they produce.
