Comment on ‘Sputnik moment’: $1tn wiped off US stocks after Chinese firm unveils AI chatbot
artificialfish@programming.dev 5 days agoNah, o1 has been out how long? They are already on o3 in the office.
It’s completely normal a year later for someone to copy their work and publish it.
It probably cost them less because they probably just distilled o1 XD. Or might have gotten insider knowledge (but honestly how hard could CoT fine tuning possibly be?)
leisesprecher@feddit.org 5 days ago
Deepseek showed that actually putting thought into the architecture achieves much more than just throwing more hardware at the problem.
This means a) there will be much less demand for hardware, since much more could be run locally on regular consumer devices. And b) the export restrictions don’t really work and instead force China to create actually better models.
That means, a lot of the investments into the thousands of AI companies are in jeopardy.
bobs_monkey@lemm.ee 5 days ago
Realistically, the CCP is probably throwing a lot of money at developers to get something good going and available, and US companies are whining about how it’s not fair. The fact of the matter is that a solid product is available for much cheaper, and US companies are now screaming foul. Guess what, a superior product made of good code (people) beats out just throwing money at hardware, who’d’ve gone an thunk it.
leisesprecher@feddit.org 5 days ago
What I find really fascinating here is that obviously openAI, Meta, etc. seem to be structurally incapable of actually innovating at this point.
I mean, reducing training costs by literally an order of magnitude just by writing better software is astonishing and shows how complacent the large corporations have gotten.
artificialfish@programming.dev 2 days ago
Meta? The one that released Llama 3.3? What are you talking about?
algorithmae@lemmy.sdf.org 4 days ago
You can write off hardware purchases, paying for skilled devs is like pulling teeth
artificialfish@programming.dev 2 days ago
I think “just writing better code” is a lot harder than you think. You actually have to do research first you know? Our universities and companies do research too. But I guarantee using R1 techniques on more compute would follow the scaling law too. It’s not either or.