FuckButtons

4 ms·

61.
by FuckButtons 3mo ago
One presumes that in order to compete they must also be innovating, if the only innovation they had was how much money they could set on fire then OpenAi / Google would have walked all over Anthropic by now.
62.
by FuckButtons 4mo ago
… by some fool who didn’t know what they were doing, since evidently the same behavior could be done with less.
63.
by FuckButtons 4mo ago
It’s not like this is a new tactic, china has been very successful at wiping out competition by undercutting them using state subsidies in many sectors of the economy.
64.
by FuckButtons 4mo ago
They already have those, for several years if I’m not mistaken.
65.
by FuckButtons 4mo ago
I’m not sure how you construe “country implements desperate measures during time of existential crisis” with technology only working for a small cabal? If anything this shows that this particular technology is more egalitarian than its pred
66.
by FuckButtons 4mo ago
but we’re going to get a 90% cost reduction in the next 18 months… right? Right guys? Sam Altman wouldn’t lie right?
67.
by FuckButtons 4mo ago
Laying the blame for the transparent financial manipulation we are observing at the feet of regular people (who are putting their savings into their pension funds, a system that we incentivize because of its pro social outcomes) and saying
68.
by FuckButtons 4mo ago
Sure, but you lost the non technical audience by the end of the first sentence.
69.
by FuckButtons 4mo ago
Some of the exams in Berkeley were brutal, but they never felt like trick questions, they did on occasion require a level of mastery of the material which was extreme, but it never felt like someone was just trying to make the questions obt
70.
by FuckButtons 4mo ago
No one consented to training llms, as the op clearly implies, if they had been asked they would have declined to do so. As would all of the many copyright holders who are in the process of suing the model companies.
71.
by FuckButtons 4mo ago
They all know it’s coming, if it pops before they ipo then they don’t get their billion dollar payday, they have every incentive to move quickly.
72.
by FuckButtons 4mo ago
Well, you are learning something, just the thing you’re learning has an even shorter usable lifespan than programming languages, namely you’re learning what works to get useful responses from ai agents. Whether or not that has value to you
73.
by FuckButtons 4mo ago
They only showed the benchmarks where they outperformed?
74.
by FuckButtons 4mo ago
> Well, AI costs are definitely going to go down at least 90% in the next ~18 months for the same quality of output (and probably 90% again in the 24 months after As far as I can see, token costs have been steadily increasing over the pa
75.
by FuckButtons 4mo ago
These are fundamentally different points in design space though, hbm doesn’t have a 10mw idle draw like lpddr does.
76.
by FuckButtons 5mo ago
Lotr ~1.4k pages 1 page ~1k tok -> ~700k tok 1tok ~0.5J -> 350kJ ~100Wh ~1 cup of tea
77.
by FuckButtons 5mo ago
Based on some napkin math, that would be about ~100 watt hours of electricity on an H100 cluster, or, roughly the same amount of energy needed to boil a kettle for a cup of tea.
78.
by FuckButtons 5mo ago
I do wonder if a wavelet transform might be better.
79.
by FuckButtons 5mo ago
You assume that the people who are at the top of the organizations generating said wealth will have any incentive to do that. Look around the world at the petro states for examples of a highly capital intensive industry generating money tha
80.
by FuckButtons 5mo ago
The problem is, it’s not like being a pm or manager, the people you work with generally aren’t pathological liars with severe amnesia. The anxiety you feel is because there is a system that is very difficult to control reliably injecting en
81.
by FuckButtons 5mo ago
But the people who want to do local inference are putting some amount of value on privacy that’s not captured by the raw monetary value so just comparing the price is somewhat beside the point, it’s also true that, if you have eg a Mac and
82.
by FuckButtons 5mo ago
It’s shocking how close this feels to claude, obviously it's much slower, but I don’t know that it’s significantly dumber. Interestingly the imatrix quantization seems to be better than whatever quant the zdr inference backends on open
83.
by FuckButtons 5mo ago
Sure, ETA for a nuclear power plant is ~15-20 years from now, wind is ~5-10, solar is 2-6, natural gas is 2-4 years. Data centers take 18 - 24 months. Even if build out for power infrastructure needed had started when the demand became appa
84.
by FuckButtons 5mo ago
Epoll / select? since everything is a file, you can wait on everything.
85.
by FuckButtons 5mo ago
Electricity usage in N.America / Europe has been static for the last ~20 years.
86.
by FuckButtons 5mo ago
Sure, but, given that, does it not seem like the conclusion is: if you have something that could in principle be reverse engineered by a competitor with more compute, they can and will steal it, because the only constraint is roi.
87.
by FuckButtons 5mo ago
I think most people who are into c that I’ve met quite like header only libraries. Copy paste as a package manager does have its benefits.
88.
by FuckButtons 5mo ago
Because those are very capital intensive and don’t skew towards germanys existing competitive advantage in diesel engines and high precision heavy engineering. Same reason most places don’t try to compete, it’s cost prohibitive to do so.
89.
by FuckButtons 5mo ago
Not if all human labor becomes surplus to requirements.
90.
by FuckButtons 5mo ago
Single interaction? Buddy that’s my entire life.

More ›