make3

16 ms·

181.
by make3 6mo ago
1.5B and insanely under-trained. Size didn't mean much before the scaling laws papers came out.
182.
by make3 6mo ago
we essentially have AGI right now brother
183.
by make3 6mo ago
there's no due process in the US, the state is completly corrupt and captured
184.
by make3 6mo ago
it's a bit scary how a lot of people are completely fine with an insanely wrong status quo of a fully corrupt and anti democratic government and of ppl like Altman allying with them.
185.
by make3 6mo ago
not in the UK (or the US) since king Henry VIII made himself Pope of his new Anglican Church in the 1500s
186.
by make3 6mo ago
the thing could barely make full grammatical sentences, it's funny to see that even then they were overclaiming the fuck out of their myself
187.
by make3 6mo ago
either that or AAA games mostly becoming streamed from Clouds like Nvidia now
188.
by make3 6mo ago
gaming too I guess still
189.
by make3 6mo ago
Trying to build things in a way that they have to be resistant to missile and drone stand is not a very reasonable scaling constraint. That's why we need non stupid leaders who don't start pointless wars
190.
by make3 6mo ago
hitting the electrical grid is a war crime, not so much an AWS server farm. optics / progressive escalation does matter
191.
by make3 6mo ago
it's like a faster than light spaceship company using horses. There's been infinite solutions to do this better even CPU only for years lol.
192.
by make3 6mo ago
they also need to understand human behavior a little bit I guess, & that's pretty complicated (so not just social, social of humans specifically)
193.
by make3 6mo ago
it's funny because domesticated cats have much more developed frontal cortexes than their ancestors & it would be one of the things that feral cats lose to genetic drift (meaning, no conservation pressure in the wild). whatever bor
194.
by make3 6mo ago
Absolutely not, the former relies on controversial ideas to qualify as legal. Stealing the content from the whole planet & actively reducing the incentive to visit the sites without financial restitution is pretty bad.
195.
by make3 6mo ago
but doesn't it break the assumption that it should ideally be able to reproduce your original results
196.
by make3 6mo ago
why me specifically or only though. your comment comes off as peachy and annoying.
197.
by make3 6mo ago
just "eat the food you cooked" would be fine. drink your own champagne sounds like such corporate group think bullshit I'd be tempted to change workplace on the spot
198.
by make3 6mo ago
tanks must represent like 0.001% of fuel consumption lol Road uses like cars, trucks and buses is 47% of all oil, and clearly an enormous fraction of that can be converted to use electricity instead
199.
by make3 7mo ago
It's funny talking about non software stuff on HN. I'm sure there's hundreds of papers on simulations and expert analysis of this.
200.
by make3 7mo ago
read up on the RNA world theory, it's so cool
201.
by make3 7mo ago
why is someone making money out of it a bad thing, if they're not preventing other people from using the code?
202.
by make3 7mo ago
I don't know why it's still a respected brand when everything they do is mid at best
203.
by make3 7mo ago
I always wonder why rockets are millions of dollars each, that seems insane to me.
204.
by make3 7mo ago
spending your productive output making one of the most powerful terrible people alive more powerful just sounds depressing to me. do you really want your research to be how to make the personality of Elon's anti woke mind virus LLM be
205.
by make3 7mo ago
they all made 600k-800k USD and up and are highly employable, you definitely can blame them
206.
by make3 7mo ago
It was up last week, now no one really needs to read the post
207.
by make3 7mo ago
You could try to get a degree on the side (not saying necessarily in comp sci) just to make your life more resilient to bad economic situations, and maybe more interesting
208.
by make3 7mo ago
insane cope
209.
by make3 7mo ago
It's weird to me to train such huge models to then destroy them by using them a 3 bits quantization per presumably 16bits (bfloat16) weights. Why not just train smaller models then.
210.
by make3 8mo ago
I would never use Slack by OpenAI lol

More ›