AYBABTME

8 ms·

61.
by AYBABTME 2y ago
Locality of data and computation is very important in neural nets. It's the number one reason why training and inference are as slow as they are. It's why GPUs need super expensive HBM memory, why NVLink is a thing, why Infiniband
62.
by AYBABTME 2y ago
Inference is not where the majority of this CAPEX is used. And even if, monetization will no doubt discourage developers from dispensing the secret sauce to user controlled devices. So I posit that data centres inference is safe for a good
63.
by AYBABTME 2y ago
If the hardware can be used more efficiently to do even more work, the value of the hardware will hold since demand will not reduce but actually increase much faster than supply. Efficiency going up tends to increase demand by much more tha
64.
by AYBABTME 2y ago
You know it's the same everywhere? It's hopeless to wish for all of humanity to change their common intuitions and independently reproduced heuristics. I'm white and spend a lot of time in Korea. I can get around in Korean. D
65.
by AYBABTME 2y ago
Oh sure, I do apologize, just wanted to know if the numbers came from somewhere or if they were a wild guess. Thanks for bringing up the references. The difference is drastic indeed. I almost would qualify for Jeju Air.
66.
by AYBABTME 2y ago
The more likely scenario to me is that they attempted a go-around after the bird strike and failed, leading them to land with gears up, no flaps, at high speed.
67.
by AYBABTME 2y ago
Are you taking that number out of your arsenal or from a source? AFAIK most South Korean pilots are actually doing their training in the US, since it's much easier to accumulate the hours, and there's more GA. I would think their
68.
by AYBABTME 2y ago
Manufacturers made planes that didn't need it, making their plane's TOC more attractive.
69.
by AYBABTME 2y ago
Seems quasi-impossible on a commercial airliner. This thing is going to beep and alert like hell about the landing gears being up. The most likely cause is a complex series of event that led to the landing gears not coming down. Landing gea
70.
by AYBABTME 2y ago
There seems to be Russian ships accidentally sinking as well, these days.
71.
by AYBABTME 2y ago
But 20y isn't ancient, as far as I can tell?
72.
by AYBABTME 2y ago
I came to the same conclusion about Temporal for these types of things. Interactive stuff that touches 1 DB? Do it in the API. Needs to coordinate >1 thing? Temporal. Orchestrating a bunch of LLM calls is a perfect fit for Temporal.
73.
by AYBABTME 2y ago
I found that this ends up being a giant amount of useless code, and a ton of memory allocation noise, that only satisfied my desire for elegance. I've given up that approach and just use protobuf types throughout as the base type. I go
74.
by AYBABTME 2y ago
Even on a high quality screen, it's a bit much.
75.
by AYBABTME 2y ago
Why not?
76.
by AYBABTME 2y ago
Consequence: we'll find life wherever we go, and we'll never know if it was always there or if we brought it there?
77.
by AYBABTME 2y ago
It can be slower than pure Go options in some scenarios.
78.
by AYBABTME 2y ago
Example? Curious, can Zig can be used to simplify CGO dependency management in general, or only for Zig programs?
79.
by AYBABTME 2y ago
If I were that guy, I'd go all in on FBTC or some ETF. Keeping a stash of distilled gold in your house is a huge liability.
80.
by AYBABTME 2y ago
Yeah, then get the double masted one? They made huge sacrifices and redesign to focus on a feature they barely use. Which is why I think it's silly.
81.
by AYBABTME 2y ago
People aren't surprised that crappy overloaded rafts sink. But they're surprised that top-of-the-line yachts sink in a minor squall. It's not confusing to me why one is more noteworthy than the other, technically. The wealth
82.
by AYBABTME 2y ago
"Single mast for better sailing performance" and then "only sailed once in 5 days, motored the rest" silly tradeoffs.
83.
by AYBABTME 2y ago
7k$ is about what yards quoted for such a job in NorCal. It's easy to DIY if you know what to do, but (1) it's a nasty job (2) it's toxic (3) it takes time and (4) you still end up paying about 2k$ if you're thrifty (in
84.
by AYBABTME 2y ago
You can undo 50 years of research easily, by this process which we call science.
85.
by AYBABTME 2y ago
I did, and I feel like only long and cold headed analysis brought me out of it. It's easier to believe that alarmism is warranted than to think it isn't. The entire public sphere is freaking out about everything all the time, and
86.
by AYBABTME 2y ago
FT does much better.
87.
by AYBABTME 2y ago
(1) might be both common and also wrong. (1) being why I dislike most of these media outlets. (2) the editorial board may be wrong and suffering groupthink, and ultimately slowly killing the paper.
88.
by AYBABTME 2y ago
I don't think it's a bad business decision. I stopped reading the WaPo when I got sick of its partial treatment of everything. And I'm not talking about supporting one candidate or another, I'm talking about sticking to
89.
by AYBABTME 2y ago
Think of the summary of a zoom call. Or of a chapter that you're not sure if you care to read or not. Not all content is worth consuming, and not all content is dense.
90.
by AYBABTME 2y ago
You haven't used them enough. Everytime an LLM reduces my search from 1min to 5s, the LLM pays. Just summary features: save me 20min of reading a transcript, turn it into 20s. That's a huge enabler.

More ›