Cacti

14 ms·

91.
by Cacti 3y ago
Ok that’s a fair point. You can still do small batch, but I agree, the results wouldn’t be anymore more than what you’d get from a laser printer and a glue binding.
92.
by Cacti 3y ago
:) I think maybe the joke was a hair too subtle for this crowd.
93.
by Cacti 3y ago
I’m not criticizing them or the approach. That’s what I would have done most likely. But the things you mentioned aren’t particular to stable diffusion, or even VAEs. Yes, the best way to learn is to be told or to build up applied/impl
94.
by Cacti 3y ago
All your points are good ones and were knowable by any researcher at the time who wasn’t, idk, a new grad or new to CV. I always assumed they just threw the VAE in there using the default options from the original VAE paper and never though
95.
by Cacti 3y ago
Oh here we go again.
96.
by Cacti 3y ago
you want to make kickstarter an international bank?
97.
by Cacti 3y ago
You can easily print far, far smaller batches. That’s not the issue.
98.
by Cacti 3y ago
lol what normal nervous system
99.
by Cacti 3y ago
frankly you can’t really trust any history. we all are the heroes of our own story.
100.
by Cacti 3y ago
I’m confused here. Are you asking if I would get in a plane piloted by someone with zero experience or training?
101.
by Cacti 3y ago
This is precisely what most safety researchers were asking for in 2016 when openai was recruiting, and why many didn’t go to openai. Like, there’s a lot of other security and safety researchers out there. The OpenAI types draw from an actua
102.
by Cacti 3y ago
Is this a boob joke or a money joke?
103.
by Cacti 3y ago
I know why uninformed comments pop up here on non-tech articles, but why is it always so aristocratic and elitist. Was the bullying really that bad?
104.
by Cacti 3y ago
The why is kind of incredibly important though in law.
105.
by Cacti 3y ago
well that’s arbitrary.
106.
by Cacti 3y ago
Oh. I understand. Carry on then lol Been there. And you’re right.
107.
by Cacti 3y ago
You let newcomers dick around for months with a net on an industrial vision problem? This stuff was solved two decades ago. Why didn’t you just tell them?
108.
by Cacti 3y ago
there are uncountable sets all over the place, and in practical terms, the repl loop may have a week long training lag after you hit enter. also, the data is almost always complete shit. lol. there’s no mystery why it’s hard.
109.
by Cacti 3y ago
I have done ML R&D for I guess 7 years now. It doesn’t get easier. You just get used to it.
110.
by Cacti 3y ago
well, ar is going to be the real money maker by probably several factors.
111.
by Cacti 3y ago
yeah. people really have a hang up on VR and AR, but it’s almost entirely simply because the electronics and power aren’t there yet. once it is, it will almost already be game over, because… i mean literally everyone eventually will have on
112.
by Cacti 3y ago
what are you talking about
113.
by Cacti 3y ago
For everyone who hates YAML, we extend YAML to another use. It’s like without rules none of you show any common sense. Who cares what the spec says? Obviously you shouldn’t use “oN” as boolean true.
114.
by Cacti 3y ago
There’s lots of crap on Arxiv.
115.
by Cacti 3y ago
AI is not intelligent, even in the abstract sense
116.
by Cacti 3y ago
20 years ago most of us would be dead.
117.
by Cacti 3y ago
no, it won’t
118.
by Cacti 3y ago
Problems reducible even partially to matrix math are for many practical purposes embarrassing parallel even within a single core. A couple hundred million FLOPS with 1990s SIMD support will let you run nearly all near-SOTA models within, id
119.
by Cacti 3y ago
thank you for the post and good work. can I ask, is the focus primarily on inference? is there anything serious going on with training at the power scale you are talking about?
120.
by Cacti 3y ago
What are the numbers for production scale (aka above craft)? Like, a thousand? A million?

More ›