Thread 105626396

how long will this ai meme bubble last? this is insane

https://x.com/Yuchenj_UW/status/1935116041866330378

>ai talent
new retarded ai buzzword

>language models are generating cognitive debt, people do not know how to do things now.
>language models generate unmaintainable code
>language models pollute the ecosystem
>because the current internet is saturated with synthetic data, the new trained models are of inferior quality, because the data pools are contaminated with generative artificial slop

>>105626396 (OP)
LINE GO UP
LINE NEEEEEDS TO GO UP
PLEASEPLEASEPLEASE INVEST GODDAMNIT!!!
SHIT!!!!!

>>105626396 (OP)

This is obviously propaganda to boast OpenAI since they can't deliver on the promistes they have made and if they don't figure shit out soon, VC won't continue to finance their failures.

>>105626706
It's worse than that actually, this has nothing to do with boasting for OpenAI. The podcast is his brother's podcast and only had like 3,000 subscribers before this was posted. He's doing the usual Jewish trick
>family member wants to cash in on easy goyim money
>okay bro make a podcast and I'll bring the cattle your way
>make completely outrageous unsubstantiated claim
>do it on your brother's shitty podcast because his only marketable factor is being the brother of Sam Altman
>total lie gets posted all over the internet therefore boosting the reach of said shitty podcast
>another talentless Jew now has an audience of retards that will continue clicking whatever garbage he posts
>Sam will now write in his diary that his brother gave him a blowjob in exchange for making him famous

i think as a worked you'd be crazy not to take $100m

>>105626396 (OP)
What does "AI talents" even do? Find ways to lock down info and usability for the goyim?

>>105626745
PhD Mathematics, I assume.

>>105626796
cant think of anyone more useless

>>105626805
Damn that's crazy, haha.

>>105626396 (OP)
What pisses me off is how fake and gay this is. The media is constantly hyping AI up into being something that is causing a way way bigger impact.

AI is dumb. The bots fucking up causes a huge problem to companies and they constantly have to go into damage control.

A company that I work for fell for the meme of firing chat support people to replace them with bots. Even with a lot of tweaks and teams working like crazy to make them work properly, customers are unpredictable and will always say or do something the bot can't handle, causing the bot to hallucinate.

A client actually tricked a bot into telling them they ate in fact eligible for a 4yr plan for free as compensation for their troubles. The company had to oblige.
After all the trouble, they just hired people again. And this is happening everywhere.

AI is nice and all and can be useful, but that's it, it's useful. Not this huge game changer, very far from it.

>>105626396 (OP)
Bubbles rise and fall at the command of banks/ the federal reserve.
Banks like JPM are “member banks” of the federal reserve for example.
The AI bubble will fall when they feel ready to implement policy changes, because that’s the entire point of making a bubble.
To have /yourguy/ in the hotseat at the point of chaos of everything blowing up so you can get your agenda pushed through.

Though I’m not exactly certain if it would work this time around.
When the market shit itself over trump’s tarriffs the resounding response from most of the public was “good, fuck stock baggies”
There’s too little involvement from the general population in the stock market for them to be concerned for a mostly rich fag bubble popping.

>>105626727
it's a signing bonus with some gay conditions for sure
plus you will be known as the $100m engineers to everyjeet at m$
wouldn't possibly want that target on my back tbqh

>>105626745
The user bots are completely irrelevant here, they don't care about improving them, to the contrary
This is all about getting the mythical keys to hell
Obviously the guy who spent billions on making the metaverse happen will spend even more on making ASI happen

If you studied ML and graduated before 2020, you are now easily a multi-millionaire. It was pretty rare back then.

Like it still does not take too much. If you are able to understand and implement stuff from current bleeding edge research papers on LLM topics, then you are a millionaire in the making. No PhD needed.

>>105626396 (OP)
I think AI will continue to be at its best when it has very focused use cases that it's actually suited for, and ateam of skilled handlers to make sure it's outputs are actually useful. Like the protein folding thing scientists are doing. It is another tool, not a replacement, out even convincing imitation of a human.
The more these big tech companies try to chase "general intelligence" the more problems they're gonna run into, because their goal is so unclear. What is the use case of general intelligence? Does AI having autonomy and personality actually make it more useful in any way? Isn't the whole idea of AGI just bloat?
It's simply not efficient or even really useful to try to make a machine filter through the world's entire collection of knowledge to provide answers to a question, especially when a search engine would be able to answer faster and without consuming jiggawatts of power to do so

>>105626396 (OP)
>AI talent

>>105626396 (OP)
I have AI talent saaaar.

>>105627355
>there are no rs in a strawberry
>9/11 is bigger than 9/9
>ehhh? i dun get it
>yemen. no i will not count how many that is

>>105626727
There's no way this is real, $100m is a batshit amount of money to be throwing around even in stock. That's 5x Cuckerberg's annual comp.

>Raises rent to 99 million a year
Landlords win again

>>105627043
Unfortunate that it's an insane mathchad-only field. Meanwhile the rest of us get to play musical chairs every year.

>fake and gay
>yeah we gives the slaves more money than many CEOs
No slaves CAN NEVER MAKE MORE THAN THE CHOSEN

ai is text when you think about it

>>105627535
I agree it's extremely far fetched, but if it is true we're talking like there's maybe 5 potential candidates who would be eligible for a $100m bonus and they're world renowned experts. 100s of others may be eligible for a signing bonus of 1 or a few million, etc typical bell curve. For the few $100mers just the PR of having them working at Meta is worth a pretty penny to Meta, the shit they can make that guy say like a puppet, endorsing Meta's research etc. And then there's the fact that by poaching that guy, it's not just that Meta gets him but also that OpenAI loses him which is a big deal. I don't think it's impossible, but it's doubtlessly greatly exaggerated for clout

>>105627509
Obsessed.

$100M is the total signing bonus pool for all the new hires in an year.

>>105628549
Name one world renowned expert that would be worth paying $100m for AND works at OpenAI.

>>105626396 (OP)
That's total bullshit by the way. The sign-on bonus is 100k, not 100m, and it's not upfront byt rather 4 year 'vested' eith 1 year cliff

>>105626796
phd math are clueless retards and never do anything but wank to themselves. Only phd cs may apply. The only relevant figures with math degrees have math degrees only because cs didn't exist yet back then.

>>105627043
I have a phd in deep learning from the mila My master's was.2014 then I did a stint in industry but without a phd the glass ceiling was real. Haven't had a job since.

>>105626396 (OP)
AI is just more complex "if i write this you do that", right?

>>105628902
Yes, but with statistical decision boundaries and more like 'output' than 'do'

>>105628683
moot

This is a battle for the domination of the next century like when the USA and USSR rushed to exfiltrate the nazi scientists before the other.

>>105626396 (OP)
2 more weeks

>>105626796
This. Especially PhD Maths or CS who specialize in LLMs.
>>105626805
They are not good at anything except LLM architecture. They are the only ones who actually know how the math behind it all works.

>>105627535
I read an article on The Register yesterday that Mark Zuckerberg is personally approaching top AI people and offering them $10m/year. And all they knew is that it’s “an eight-figure sum”, so it could be $99m for all they know and this could be the same story.

https://www.theregister.com/2025/06/13/meta_offers_10m_ai_researcher/

>>105630595
>They are the only ones who actually know how the math behind it all works.
lol. lmao even.

>>105626845
>4yr plan for free as compensation for their troubles. The company had to oblige.
Total bs, how would the customer prove that he received such offer? (Asking so I'm gonna go around tricking chatbots in giving me stuff for free)

>>105630800
The register is about as reliable as the onion, but they think they're serious

>>105630894
Taking a screenshot of the chat log.
At least in my country, a company is forced by law to fulfill a business agreement, advertisement etc including what a representative told them through official media. The bot is not a person, but neither the law or the client cares about that.

So let's say there's a banner somewhere saying "buy our plan for 100$" and there is no message saying "valid until X date" someone can take a picture of it and demand to buy that product for that price. If the company declines, they can be sued. So for the company, it's best to avoid the lawsuit and just do what the customer wants.

Of course, business make sure that something like this is rarely a problem. Then comes AI and how completely unpredictable and easy to trick it is. Someone with experience dealing with AI can easily fool the bot to say something bad.

In that specific case, the client took a picture of the website containing the 4yr offer using their phone, while keeping their cursor over a word, while the next word was "free".
This made the AI believe the 4yr plan was indeed free for new users.

You can't really prove that this was done in foulplay, so that's that.

>>105631299
>ai chatbot with vision capabilities
yeah that company is stupid.

>>105626845
What puzzles me most is that rule-basws systems do not hallucinate. The reason we couldn't keep using them is that you needed 50 field experts to help out fill in rules for 3 years to get a useful system.
Fast forward to today. LLMs are, in principle, the perfect tool to generate rules from e.g. specialized encyclopedia and humans describing facts in plaintezt (the bottleneck was coding rules, not writing down facts).
I have seen 0 attempt at doing this.
Or at using any of the other technologies along those lines. Basically, their stremgths are ignored and they're shoehorned where they suck most. Why?

>>105631446
Images are necessary for better troubleshooting wtf the client is talking about.
>"this button isn't working, fix this!!"
What button?
>"the OK button" (there are 8 of them in the page)
which one sir?
>"the broken one. Fix this already!!"


A human can instantly understand that this client is stupid and talk to them in a way where this interaction will go somewhere. AI cannot, they can understand English, code language etc, not moronic.

>>105631530
It's kinda what I said above. No AI can properly account for human error and stupidity. It would need to be trained and programmed for hundreds of years to account for everything a human can say and do, all the way from a moron with 10iq to a smart guy trying to fool the system. And I'm not even talking about the moral and legal issues of programming something that takes into account the client is a dumb jeet with broken english, or how the client can instantly recognize this is a bot and demand to talk to a real person (oops, we fired like 80% of them and there's a huge queue)

So yeah, leave online support for jeets.

>>105626396 (OP)
You listen to him talk and he sounds like a 16-year old girl
> so, like, um, you know, um, like

>>105626396 (OP)
Until retarded investors realized that AGI will never happen.

>>105627535
>>105630800
I'm surprised this didn't exist before. If you think football stars get similar offers between clubs, why not top scientists between companies? Both are revenue generators are the end of the day.

>>105627535
They spend 100x that on the hardware every year.

>>105632277
Investors only care about revenue. If meta launched a completely retarded LLM that people pay lots of money for, the Investors are happy.

>>105626396 (OP)
>jew whines about getting outjewed
many such cases