Sorry, but that's an exceptionally unimpressive article. The crux of his thesis is:
>The main flaw is that this idea treats intelligence as purely abstract and not grounded in physical reality. To improve any system, you need resources. And even if a superintelligence uses these resources more effectively than humans to improve itself, it is still bound by the scaling of improvements I mentioned before — linear improvements need exponential resources. Diminishing returns can be avoided by switching to more independent problems – like adding one-off features to GPUs – but these quickly hit their own diminishing returns.
Literally everyone already knows the problems with scaling compute and data. This is not a deep insight. His assertion that we can't keep scaling GPUs is apparently not being taken seriously by _anyone_ else.
Was more mentioning the article about the economic aspect of China vs US in terms of AI.
While I do understand your sentiment, it might be worth noting the author is the author of bitandbytes. Which is one of the first library with quantization methods built in and was(?) one of the most used inference engines. I’m pretty sure transformers from HF still uses this as the Python to CUDA framework
Scale. You can monetize on the people that don’t pay the $200/month. Obviously I have nothing to prove this statement, but I wouldn’t be surprised if the subscriptions are loss leaders.
Not OP but I doubt it. I’m in my mid 30s and when I grew up there in the 90s, Suffolk county was bumble. Some people had horses on their land. After 9/11, a ton of people moved in from the city and the population absolutely ballooned. Over two decades, the population grew so much that just Nassau county and Suffolk county combined has more people than a handful of states. People come and go too (including myself) so unless some organization is tracking us, it’ll be hard to pinpoint.
Aha likewise, I swear, between the ticks and the polluted water, a good amount of us are screwed. Grumman has put some nasty stuff into the ground too. I remember growing up how they mentioned it was slowly seeping into the aquifer. Took me ages to convince my parents to get a RO machine
I worked at BNL during college days through the SULI program! Some of my peers from college is working there full time now too. I got to work on some really cool stuff but unfortunately a lot of the tenured researcher I knew have seem to left. I heard a lot of researchers left during Trump’s first term.
First time hearing the term micro bakery. Is there an equivalent for restaurants? Like a micro restaurant? Seems like a fun idea for my retired parents.
As a MechE turned SWE, always a fun read when SWE try hardware.
> Blink and you’ll get a different measurement.
This means your environment is not controlled enough. Also quality control is usually done in terms of statistics. You might want to read something called gauge R&R. That being said, you should be proud of being able to ship a physical product!
As for quality checks, software quality teams pales in comparison to hardware quality teams. Mainly as you said, there’s a lot checks you can do in software. For hardware, bigger companies have to have their vendors qualified. The vendors have to follow their customer guidelines and do outgoing inspection. Then the company has a division to do incoming inspection. There’s a traveler that follows the kit (of parts) and there’s usually subassembly quality checks. Then final full build checks before it leaves the door.
> terms of statistics. You might want to read something called gauge R&R
I think the first thing to focus on is the stats portion - do you have appropriate FAI/SPC/OQC with Cpk requirements defined? Gauge R&R plays a much smaller role, especially in something that is relative
reply