Inside Meta's AI Drama: Internal Feuds Over Compute Power (theinformation.com) 28
An anonymous reader shares a report: Meta Platforms' releases of its large-language models, Llama and Llama 2, in the past six months have won the company praise for offering free, open-source alternatives to models from OpenAI and Anthropic. But for some of the scientists and engineers who worked on Llama, that praise was too little, too late. Many have quit, embittered by a previously unreported internal battle over computing resources with another Meta research team working on a rival model that the company ultimately abandoned, according to people with direct knowledge of the matter.
Their departures are a reminder of the simmering tensions over shortages of computing power that both big and small tech companies are grappling with in developing generative AI, which requires specialized chips that aren't freely available. Big tech firms have more computing resources than most, and some, like Meta, highlight that fact when recruiting AI researchers. But even they are limited in what they can supply. The exodus also highlights the challenges for big tech companies in retaining highly prized AI researchers as demand for AI talent surges. More than half the 14 authors of the original Llama research paper published in February have left the company, several for AI startups or other big companies, the people with direct knowledge said.
Their departures are a reminder of the simmering tensions over shortages of computing power that both big and small tech companies are grappling with in developing generative AI, which requires specialized chips that aren't freely available. Big tech firms have more computing resources than most, and some, like Meta, highlight that fact when recruiting AI researchers. But even they are limited in what they can supply. The exodus also highlights the challenges for big tech companies in retaining highly prized AI researchers as demand for AI talent surges. More than half the 14 authors of the original Llama research paper published in February have left the company, several for AI startups or other big companies, the people with direct knowledge said.
So, limited resources are limited (Score:1)
Pompous assholes arguing over resources?
Maybe look around a bit?
It happens every day all over the planet.
Re: (Score:2)
"previously unreported internal battle over computing resources with another Meta research team working on a rival model.
A little internal camaraderie in the form of healthy competition is one thing, but this sounds more like a couple of teams literally aligned against each other UFC style within the same company. Go figure you just might sustain some employment casualties if you let that competitive drive get out of hand.
In a world infected with mass narcissism, I just can't imagine how workplace arrogance might have gotten out of hand. It's almost as if there's a downside to social media addiction.../s
Re:So, limited resources are limited (Score:4, Informative)
"previously unreported internal battle over computing resources with another Meta research team working on a rival model.
A little internal camaraderie in the form of healthy competition is one thing, but this sounds more like a couple of teams literally aligned against each other UFC style within the same company. Go figure you just might sustain some employment casualties if you let that competitive drive get out of hand.
In a world infected with mass narcissism, I just can't imagine how workplace arrogance might have gotten out of hand. It's almost as if there's a downside to social media addiction.../s
In some cases, business units or departments are pitted against each other, full stop, all out war, winner take all style by management as a "healthy way to develop more aggressive business strategies." I literally was told this by a previous company CEO when asking why it felt like my department was at war with another when we should all be working toward the same end-goal. It was literally by design. And the higher ups were even placing bets on which team would win the intentional war of aggression.
Super healthy.
Re: (Score:2)
Again the emphasis on healthy competition means business strategies will appear out of the ether. Just as in the real world. In the end, employment should not suffer casualties as you have room to absorb the "losing team" to other areas of your business.
Of course as last year showed, entire teams will just get chopped because management sneezed and as such competition devolves from healthy into a free-for-all using underhanded tactics to beat others to production
Re: (Score:1)
Re: (Score:2)
Capitalism within capitalism! Genius. Pit the workers against each!
We're going to lay off (what's a good round number?) 10,000 workers. Maybe another 5k tomorrow. We'll see, depends on what the first round of layoffs does for the stock price. Everybody work harder! Have a nice day.
I'd quit too - layoffs suck even if they say "don't worry, you're safe"
Corporations are central planning. The chairman, I mean chief officer, decides how to allocate resources.
Capitalism inside a corporation is stupid. Only central planning inside a corporation.
Re: (Score:2)
What's the alternative?
Humans are intrinsically resource-driven and competitive. It's a feature (bug?) of our limited-resource world. The organisms which are able to more effectively assure themselves a future survive; the others, do not.
Encouraging competitive mindset taps into this evolutionary feature and will get more results, faster.
I'm not sure how you could say it's "unhealthy". We need a certain degree of pressure to thrive.
Re: (Score:2)
What's the alternative?
Humans are intrinsically resource-driven and competitive. It's a feature (bug?) of our limited-resource world. The organisms which are able to more effectively assure themselves a future survive; the others, do not.
Encouraging competitive mindset taps into this evolutionary feature and will get more results, faster.
I'm not sure how you could say it's "unhealthy". We need a certain degree of pressure to thrive.
You've 100% bought into the narrative that there is no alternative to capitalism and now are adding your own hallucinations that humans are nothing but competitive.
Humans are collaborative as much or even more than they are competitive.
Our main evolutionary feature is that we communicate and collaborate!
Re: (Score:2)
You're mistaken - I haven't bought into the 'capitalism lie'. But go on - tell me what your ideology is while demonstrating your personal economic strata while revealing you don't understand personal motives.
Humans are collaborative and communicative - you're absolutely correct. Do you know the network size for that to be possible, no a 1:1 basis? You're looking at a population size of less than 1,000, and a personal network of 100-200.
Beyond that, you're looking at a top-down dictatorial model. And, of cou
Re: (Score:2)
Yep, yep. And, I know where you aholes go from here. You say we need to cut off all social benefits because we compete and nobody should be given anything. Everybody should compete.
But, in your own world, you benefit from nothing but co-operation of others and grift. You attribute all you success to your ability to compete but in reality its just being born with a good social network and help from others.
Re: (Score:2)
If teamwork aligned as a whole is an ineffective model, then even the individual teams model is broken. Teamwork, works. Not having to deal with the extra burden on a project worrying about the status of"them" vs. "us", can create the boost in efficiency that likely eradicates any artificial benefit executives were trying to create with stress via unnecessary competition.
Employees are competing. All the time. It's called the competition. Doubt business really needs to add more related stress and call t
Re: (Score:1)
Re:So, limited resources are limited (Score:4, Insightful)
These "teams" aren't really your usual corporate employee. They're mostly academics who have been given more money than they know what to do with and are racing to gather more fame to go with it.
It's also just a smart move to quit right after you've done something notable. The company you're at is never going to give you as good a promotion as the company you're not at. Typical white collar workers who remain at the same company make 50% less than the ones who move every couple of years. And that's the average; the ones who get public credit for the big successful products and advances are likely much worse off if they stick around.
Re: (Score:2)
These "teams" aren't really your usual corporate employee. They're mostly academics who have been given more money than they know what to do with and are racing to gather more fame to go with it.
It's also just a smart move to quit right after you've done something notable. The company you're at is never going to give you as good a promotion as the company you're not at. Typical white collar workers who remain at the same company make 50% less than the ones who move every couple of years. And that's the average; the ones who get public credit for the big successful products and advances are likely much worse off if they stick around.
Are these the same academics who burn most of their 20s getting a PhD while their peers get six figure salaries and six figure stock options? When they have their phd their age group peers are millionaires?
Re: (Score:2)
Yes. Then some company comes along and offers them a seven figure salary and promises of resources beyond their wildest dreams to pursue their pet projects. Oh, and you can still publish like a regular academic too.
Re: (Score:2)
I hope you don't have climate crisis in the alternate reality that you're from.
Re: (Score:2)
Ah Slashdot. The weird-ass random replies are what makes it all worthwhile.
Re: (Score:2)
What? Are you badmouthing the dogma of perpetual growth and unlimited resources?
Heretic! Burn him on a pile of stock options!
Llama Team Exodus isn't True (Score:5, Informative)
More than half the 14 authors of the original Llama research paper published in February have left the company, several for AI startups or other big companies, the people with direct knowledge said.
You don't need direct knowledge of this, just check LinkedIn. Only 4 of the 14 authors of the original Llama research paper have left Meta. Three of them went to start Mistral AI and one of them was only a PhD student when the paper was published (and he works for Inflection AI now). That is much less than half. All of the LinkedIn pages I saw looked pretty well maintained, so it's doubtful they went to work for another company without updating their LinkedIn page.
I guess it's possible that another 4 Meta employees all left the company over the past few months without caring about their LinkedIn page, but I also couldn't find any press releases about any of the others leaving Meta for some startup. On the other hand the ones who left for Mistral caused big enough news even for a Slashdot post in June, so stories of others leaving would probably show up in a quick Google search.
Re: (Score:2)
Solid fact-checking work, sir.
Drama? (Score:2)
I think it's a comedy.