Yann LeCun says xAI is "kind of a failure" – and the whole AI industry might be headed for a reset
2 days ago by sanitation to c/technology
More GPUs =/= better AI.
More data =/= better AI.
More tech bro “superstars” =/= better AI
This is what people like Musk and Zuckerberg don’t seem to understand.
Training scales very poorly past a certain cluster size, especially if you go for new architectures to actually pursue improvements, hence reports of GPUs being tasked with busywork just to meet utilization quotes. Increasing data size and training scale hits diminishing returns, quick, or even regresses models because the bulk data is shit and the model is too inefficient. A prime example: Llama 4. “Superstar” AI engineers are better and Tweeting and sycophantic gaslighting than coding something interesting.
In other words, I’d argue there’s a much smaller “sweet spot” for pure LLMs that these billionaires are way, way past. And no one is telling them no because they’re too rich to hear it. It’s all going to collapse on itself because scaling like that just does not work.
This is what people like Musk and Zuckerberg don’t seem to understand.
They know, but can't admit it. Pretend and keep the stock price going up makes them money, so for them it's still working. Meanwhile they hope for a breakthrough, bail-outs or a new hype train to jump on.
No, they don't.
They're surrounded by yes men. And from everything I hear them say, they don't understand the first thing about how LLMs actually work.
AI is so weird, it absolutely is useful and has uses, but it absolutely is NOT what these billionaires want it to be.
And even IF IT COULD BE all they claim, it would literarily cause an economic collapse.
The truth is that AI is a tool that's useful, but anyone in those companies trying to dampen things will be immediately moved to the side in favor of a guy saying, dude we can go to the moon with this trust me!
So yea... I don't know how we get out of this... It's very similar to the dot-com bubble. Internet was/is important and transformative but it wasn't what everyone was claiming... So the crash came.
They don't even need to dampen, just redirect. Pay engineers to make them specialized tools, don't waste so much money on GPUs, and give part of the compute to the community to tinker with, instead of hoarding it and doing squat.
no shit sherlock
Wow, REEEEEEAAAALLLLLY, Yann? So fucking astute! We gotta get this guy to Vegas, IMMEDIATELY.
Says the person who took a billion to make the next thing after the reset.
You shouldn't respect anyone's opinion: you hear the opinion and judge the contents on it's own merit. Everyone you respect will have the odd terrible take, and people you loathe will periodically have a banger thought.
A ton of Dawkins fan boys are huffing copium since the man decided Claude is conscious (and for some incelly reason a woman). A lot of MAGA asshats are trying to torch data centers.
If you offload your own critical evaluation of ideas merely to an adjudication of the speaker... honestly I hear it no differently than "because chatGPT said".
This is just the laziest ad homium. Tell me, what do you think about the content? I am genuinely curious, and that analysis is 1000x more valuable.
I respect this opinion
3 reasons: "one of the "Godfathers of AI ... who previously served as Meta's chief AI scientist"
And you just don’t care to post any examples.
Apologies. I thought it was clear from context. Here are some examples of his bad judgement:
“one of the "Godfathers of AI … who previously served as Meta’s chief AI scientist”
Oh silly me! I thought your question mark meant you were looking for an answer, but you had already decided on an answer and only asked to find someone to vent to! I'm glad I could help you for free! Now fuck this!
It's not like meta is doing a literal holocaust. Given a good salary i'd work for them. For this guy his time at meta prolly enabled him to be more public with a rather anti corpo position. I can understand that move.
Their leadership actually made substantial contribution to enabling the Rohingya genocide.
Sure they weren't the ones burning families alive in their houses.
But IMO knowingly focusing on avoiding and limiting responsibility instead of immediately working on resolving the role of FB in the Rohingya genocide, has a level of equivalency.
This is just one example. There are many others spanning a range of areas in including fraud enablement, user harm
Btw, I am not saying I would judge everyone who works at FB. Although I feel there would be few exceptions for anyone in a somewhat senior position or above (be it an a technical or business type position).
It’s not like meta is doing a literal holocaust.
He doesn't anymore
Yeah but the association is enough for some people. He probably doesn't even use Arch Linux.
...I think people who use Linux are the exception???
Musk faces challenges attracting engineers
No shit. I have no idea who still works at Xitter. I think they must have whittled it down to workaholic Nazi pigs who personally share in Elon’s psychopathy. Because no self respecting tech workers would put up with his bullshit when there’s a whole industry they could go work in.
SpaceX is different. Aerospace is a narrower field and SpaceX is truly at the forefront of it. For a time, Tesla was also at the forefront, doing things that no one else was.
But Xitter and xAI or whatever the fuck it’s called… they have zero advantages with which to attract engineering talent. Elon doesn’t know what the fuck he’s doing when it comes to software and never has. He tries to push tech workers around the same way, he’s going to find they can get other jobs at superior organizations and likely for more money.
Since Elon Musk is a super genius who sleeps on the floors of his factories couldn't he just run xAI himself without any staff?
Dot com 2.0 wooooo
We can only hope
Posting Part 1 and Part 2 which breakdown LeCun's JEPA model and more.
@sanitation 乐观点看，LeCun 说的“failure”可能更多是指 xAI 当前的商业化进度，而不是技术本身。比如 Grok-1 的开源就证明了模型能力还在——只是产品化的 timing 和路径可能跟 OpenAI 不同。你觉得这种“技术 vs 商业”的二分法在 AI 创业里是否普遍存在？我这边观察到不少团队在开源模型上反而活得更滋润，我们整理了几个案例对比 如果对你有参考价值的话。
Wow the only technology that is designed to lie to you is a failure? Pattern recognition means it can recognize wrong patterns.
They fundamentally cannot fix hallucinations in LLMs
Why don't they just add 'don't make mistakes' to the system prompt, are they stupid?
Shit! Why didn't I think of that?
@lemmy.world
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related news or articles.
- Be excellent to each other!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, this includes using AI responses and summaries. To ask if your bot can be added please contact a mod.
- Check for duplicates before posting, duplicates may be removed
- Accounts 7 days and younger will have their posts automatically removed.
Approved Bots
go to feed...
@lemmy.world
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related news or articles.
- Be excellent to each other!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, this includes using AI responses and summaries. To ask if your bot can be added please contact a mod.
- Check for duplicates before posting, duplicates may be removed
- Accounts 7 days and younger will have their posts automatically removed.
Approved Bots
go to feed...
The Pope is kind of Catholic too.
save