Fei-Fei Li says Stanford's Natural Language computing lab has only 64 GPUs and academia is "falling off a cliff" relative to industry

these universities have endowments larger than the cash reserves of most tech companies. let’s not pretend that they can’t afford to get the resources necessary to train competitive models

maybe use some of this

Source:

@martin_casado IF ONLY THEY HAD TENS OF BILLIONS IN ENDOWMENT AND STUDENTS WILLING TO PAY CLOSE TO A HUNDRED GRAND A YEAR TO ATTEND. THEN THEY COULD SPEND MONEY ON IMPORTANT THINGS.

Our take on this issue:

What could Stanford be working on that requires more than 64 GPUs?

Academia is not supposed to be in the arm race with the industry. Heck, they should be leading the theoretical research!

Good. AI is much better off in the hands of industry than unaccountable academics.

Had not considered academic institutions in the race to AGI. What if there was a government entity like NASA but for AI…?

Wait the public sector is the innovation engine of our economy? 🤔

why would a hedge fund need GPUs?

Lol don't expect any sympathy for academics ever again.

I expected they would be using cloud compute like most tech companies apart from the absolute largest ones who own all the datacenters.

I heard @akashnet offers access to high-end GPUs to university students on an on demand basis

GPUs or administrators; Stanford made its choice.

Eventually That is all that will be required once brute force training is streamlined

cc @PeoplePlusAI @tarundua81

Between the $100k tuitions and $36B endowment Standford can afford to buy what it needs.

gpu poor builds character

64 is a lot

Stanford has almost a $10B annual budget alone. Perhaps reallocate a few dollars away from the Whitey Is Evil department, or fire one of the seven thousand diversity administrative deans and you’ll have plenty for as many GPUs as you want.

Academia is very bad in here too

@Stanford and @OpenAI should form some sort of research partnership with access to more compute. Great pipeline for intelligent future employees as well.

Me with 64+ GPUs one click away

She's a pretender.

Next we’re going to learn that the petroleum engineering department doesn’t even have their own offshore oil rig.

I am really disappointed to hear that Academia has fallen off the cliff. I was hoping to push it off myself and watch it drop to its death on the jagged rocks below.

@ddvd233 真假😰😰

This might be relevant for you. In a paper with solid data we document the gap

Sandford NLP lab has 64 GPUs according to Fei Fei but it does not mean that is what GPUs Standford has. There are many CS labs at Standrod probably.

a la atención de @joseluisescriva: vean el vídeo con calma y paciencia

(1/2) Training a highly capable AI model can quickly run you upwards of 100 million dollars while future models might end up in the 10-100 billion range - a number hardly comprehensible unless compared to the US Military budget.

Famous for being famous.