What was interesting about algorithms? The challenge of doing X with as few resources as possible (memory, power, etc). What are OpenAI + friends doing? The opposite. "Lets assume we have unlimited resources & try to build a god." No wonder they're losing their shit with the DeepSeek thing. - ThreadSky

timnitgebru.bsky.social • 27 days ago

What was interesting about algorithms?

The challenge of doing X with as few resources as possible (memory, power, etc).

What are OpenAI + friends doing? The opposite.

"Lets assume we have unlimited resources & try to build a god."

No wonder they're losing their shit with the DeepSeek thing.

Comments

raymc.bsky.social•27 days ago

It's open sourced. It'll take less than 6 months to absorb it into the rest of the industry.

dabxdee.bsky.social•27 days ago

And open AI isn't even sweating it because they're going to get that $500B and won't deliver a God damn thing. Watch.

zenturtle.net•27 days ago

this

mataliandy.com•27 days ago

My son developed several AI tools that HAD to work with the assumption that there were deeply limited resources available. He came up with a mechanism that worked.

josemarichal.bsky.social•27 days ago

There is this interesting tension between the logic of "reducing the cost function" and the shock and awe of "our model has a trillion parameters"

paulocandido.bsky.social•27 days ago

"Here here! We made this car, it can cover 1000 kilometers in only 1 hour, using a 1000 liters of gas. And no, we won't show you the engine."

"Hmm. We made a car that covers 900 quilometers in one hour, using 50 liters of gas. And by the way, here's our engine plans, you can built one yourself."

floatingoverjovia.bsky.social•27 days ago

I mean, they're also apparently not very good at it either.

thebeehammer.com•27 days ago

I think that’s the sickness of corporate American nobody ever asks “can we do the same thing easier, cheaper, and using less resources?”

The focus is on speed, not correctness.

songstars.bsky.social•27 days ago

📌

patrick23dingen.bsky.social•27 days ago

I guess they forgot there was China... They floated too much on the western world and now get a good booty whooping.

boboknowsy.bsky.social•27 days ago

US Politicians & Tech billionaires losing BIG money in stocks is hilarious. I’m here for it. 👍 And Donald thought he was doing something announcing 500 billion AI investment. 😂😂😂

aaronclow.bsky.social•27 days ago

"And Donald thought he was doing something announcing 500 billion AI investment."

I gotta say, that's the most hilarious part of this. I think some people in the Whitest House are about to go through some things.

ilogic21.bsky.social•27 days ago

Yeah of course with 95% financing lots of friends and family ...

kripasingh.bsky.social•27 days ago

It gets funnier, last year on a trip to India, Sam Altman said it's not possible for countries like India to create Open AI type solutions.

Well, he must be the cursing Chinese now.

https://www.newsweek.com/sam-altman-india-project-indus-1919694

ahfaustinomd.bsky.social•27 days ago

I'm not surprised the China is innovating at 1/10 of the cost. The US solution will be to tariff or ban it.

coffeefoxo.bsky.social•26 days ago

they saw O(n!) and said "i want that"

michaelam.bsky.social•27 days ago

Thanks Timnit!

uofagrad97.bsky.social•27 days ago

Tech Bros sadness makes me happy.

scaryscienceguy.bsky.social•27 days ago

This is why simple phone apps are 300MB long and need 1GB of RAM.

Any jr. programmer can write some poor code and it will run just fine at 4GHz with todays CPUs.

foosballwarrior.bsky.social•27 days ago

to be fair, most languages that are written in these days are not designed to be optimized. or better yet, they are so abstracted that people forget or never even know to begin with, that there is a discrete space that their code needs to "fit into" during runtime.

mcmnus.bsky.social•27 days ago

themjl.bsky.social•27 days ago

Like Cajun cuisine or the Catholic Church, Engineering has its own holy trinity: "Size, Weight and Power".

If they need a nuclear reactor to crudely emulate (maybe) 20-30% of the function of a full human brain (12-25 Watts. A dim light bulb)...?

Perhaps their current approach is not optimal...🤷‍♂️

trollingthetrolls.bsky.social•27 days ago

The censorship is problematic.

clibourne.org•27 days ago

the servers are hosted in china so this is required by law, if you were to run the model locally or on a stateside server (which you can, it's open source) it wouldn't do this

trollingthetrolls.bsky.social•26 days ago

Thanks for the clarity! Yeah it's pretty amazing to see censorship LIVE! Lol. Also would you know if there's any malicious code that can be spying or keylogging? Today my account was locked from the direct app

shenannigans.bsky.social•27 days ago

I just posted a reply showing it's the answer when you run it locally

trollingthetrolls.bsky.social•26 days ago

Ty for that!

millhouse.bsky.social•27 days ago

A lot of the tech industry feels like it's been getting sloppy over the past 40 years. Optimisation should have been the goal but brute force power to solve problems seems like the trend

agenericbody.bsky.social•27 days ago

Yes it has been getting sloppy.

Database running slowly? Throw a TB of RAM at it.

millhouse.bsky.social•27 days ago

The same in gaming. We used to have little tricks and workarounds to maintain performance. Now games are launched bloated and sloppy

timnitgebru.bsky.social•27 days ago

Addendum: whether DeepSeek or OpenAI, there's no such thing as AGI and attempting to build it results in harm.

See our paper.

richardkbethell.bsky.social•27 days ago

I don't think any existing system qualifies as AGI, but that doesn't mean it can't exist. Go back to neurology - Broca's brain doesn't do math problems, but it can express them. AGI could be designed by letting an LLM formulate a question, and ask a computational engine (ie Wolfram Alpha.)

timnitgebru.bsky.social•27 days ago

So don't make this into a "US vs China US has to win on AGI" thing please 🙄

benjaminjriley.bsky.social•27 days ago

It's definitely gonna be made into that.

freedomforall.net•27 days ago

Deepseek have fundamentally undermined the hyperscale approach, and we can all hope those malignant actors wise up and change their plans. We don't need AI slop injected into all the things. This is potentially a communal good for all of humanity, that they are reigned in by reality.

timnitgebru.bsky.social•27 days ago

Oh forgot to link our paper: https://firstmonday.org/ojs/index.php/fm/article/view/13636

gringene.org•27 days ago

"By tapping into Cold War rhetoric and framing the need to build AGI as a safety concern, TESCREALists have started to steer Western politicians and global multilateral organizations into investing in, legitimizing, and prioritizing their AGI agenda."

falschspieler.com•27 days ago

Interesting paper. I always thought AI and robotics was simply about creating servants.

badzergling.bsky.social•27 days ago

oh they do. But they want the servants to be humans servants (normal folks/working class/proletariat/etc) of the new priests (billionaire AGI dorks) to the machine G*d

eraw.bsky.social•27 days ago

This is what I’ve been looking for today to make sense of all this, thank you. It’s a eugenics cult. Of course.

doninside.com•27 days ago

it's the american way of dealing with things. Think at car engines: american ones are way bigger and fuel inefficient compared to european or japanese ones producing the same horsepower

monawys.bsky.social•27 days ago

throwing money at a problem is the first of death by a thousand cuts

kencf0618.bsky.social•27 days ago

If as if they'd never heard of ex nihilo...

dipseek.bsky.social•27 days ago

Why not have a god If we can have one?

simplekid.bsky.social•27 days ago

"Just because we can" is rarely a good reason to do something.

Why do we need one? For what purpose?

cielothelimit.bsky.social•27 days ago

😀...These were some from the 2020 Covid-shutdown times and onward, in our own Silicon Valley's suburbs. The Dwaar-Palaks, their kith, kin and pets. No brains, no decency, no humanity!

jonawebb.bsky.social•26 days ago

That's not right at all. OpenAI showed how it was possible to do much more than was thought possible for years with current resources. They've led in making their research available. And now someone else has made another advance.

Comments

Posting Rules

Reply