Online debate is zeroing in on Meta's AI slowdown after key exits and repeated restructurings, with critics linking underwhelming output to pay structures that appear easy to exploit plus high-pressure team environments that favor certain personalities over steady progress.
@eigenrobot My pet diagnosis is that the aggressive compensation formula is highly gameable and the sorts of dynamics this puts in place, and the people who can tolerate those dynamics, poisons the environment over time.
so howd meta fuck up so hard on AI
was it all just lecun's fault?
or were incumbent tech giants just structurally disadvantaged?
Aggressive packages topping hundreds of millions over multiple years coincide with reports of manipulation incentives and retention tied more to vesting calendars than delivered results.
Commentators note that Meta's long-standing open-source approach for models like Llama removed some competitive moats peers enjoy, leaving teams without the same urgency to lead closed-model races.
Many users criticized Meta, Zuckerberg, and LeCun over AI setbacks, alleged ethical violations, and leadership failures while a few praised the company's engineers and expressed optimism about future software gains.
@eigenrobot game is just starting!
so howd meta fuck up so hard on AI
was it all just lecun's fault?
or were incumbent tech giants just structurally disadvantaged?
counterpoint
and yeah fair enough
@eigenrobot Are we sure they did
@eigenrobot Same way they blew it on VR.
God-awful management.
@jeff82874662 @eigenrobot I don't have great visibility into how successful meta's native chatbots and how much revenue they drive but I can absolutely believe that there's less return to frontier models for "sexy cheerleader who helps me do my homework" bots used by high schoolers
@eigenrobot They’re using it to actually juice revenue and it seems to be working. Not clear that this is wrong in terms of a business strategy vs say, dumping a ton of spend into training frontier models with possibly dubious competitive moats
But ymmv who knows! All remains to be seen
@jeff82874662 maybe not i guess???
uncertain how it's being used internally but i don't hear much about llama in the context of frontier performance
@eigenrobot They went for open source which broke the incentives to win
so howd meta fuck up so hard on AI
was it all just lecun's fault?
or were incumbent tech giants just structurally disadvantaged?
@AlexGodofsky @jeff82874662 @eigenrobot not chatbots, they've changed their ad-ranking model to be transformer based.
@eigenrobot is it all bad I've never tried their shit but that's mostly because I'm in the back of the tech herd and no one talks about it
I guess this is a tautology, please ignore
@IE_Capital calm down
@eigenrobot I wonder if the answer is just "taking a contrary strategy made sense when lots of talent was racing the some way but that way wasnt a sure thing, but in this case the herd was right and some positive EV bets end up bad"
@eigenrobot Their failure is evident because they tried. Amazon and MS just gave up a lot earlier. In 2015, that's who we'd have guessed would be running at the frontier. Of the big incumbents, only Google has managed to catch back up to the SOTA level.
@xlr8harder yeah
i think there is unfortunately a lot of that
@eigenrobot Have you tried meta ai with meta ray bans on before
@eigenrobot wang
it's all about the money, not the tech, for them. I think they are using a LOT of ML/AI and doing pretty well, just not trying to make it a user experience.
amazon is fucking up bigtime, though. but they bascially have sold all of their brand loyalty and trust to chinese product ranking.
@eigenrobot Largely yes. He contributed to a toxic environment, pushed for open sourcing Llama (immediately creating 100 frontier competitors), convinced Zuck that test time compute would fail. Ultimately leading to Meta missing the boat entirely on reasoning models.
@eigenrobot It's just extremely hard to make frontier models in general. Even Google and xAI are behind. Without a mission or prestige no amount of money will make a frontier model
@eigenrobot >lecun's fault Lol. How fucking stupid do you have to be to ask this?