mgraczyk

47 ms·

691.
by mgraczyk 4y ago
EDIT: No
692.
by mgraczyk 4y ago
Interesting to see posts like this where people speculate on specific factual things, but they don't have any information about the specific factual thing. This post is factually wrong about how Facebook works and what the goals are. F
693.
by mgraczyk 4y ago
Not sure where you got bathroom from. The people I know who have this generally do their entire apartment. Also it gets dark here at like 5:30
694.
by mgraczyk 4y ago
I believe for internal communication, Twitter style social networks are a net bad. At Facebook we used a facebook-esque product called Workplace for this sort of thing. I believe it harmed the culture because people would chase clout and vi
695.
by mgraczyk 4y ago
In San Francisco you can spot some of these at night. Look at the tall buildings, some of the windows are illuminated with a deep red.
696.
by mgraczyk 4y ago
The video mostly seems reasonable, but the economic arguments toward the end seem off base to me. Hossenfelder describes a situation in which universities rent equipment from large companies, but fits this into a worldview in which quantum
697.
by mgraczyk 4y ago
Hmmm if you're the same srj that I think I know at Google, then it's interesting you would say that. At least amongst the managers that you and I jointly interact with (especially my own), managers seem to code plenty, maybe too m
698.
by mgraczyk 4y ago
230000 > ~5
699.
by mgraczyk 4y ago
Ok but you said "Facebook has been fined over this". You admit that's a false claim? Tiktok and Facebook both should be fined for violating the law when there is sufficient evidence that the violation happened. We don't
700.
by mgraczyk 4y ago
When was Meta fined for secretly building a system to track the whereabouts of specific individuals? You may be thinking of Lundy et al v. Meta Platforms, in which they were not fined but agreed to a civil settlement and denied they were do
701.
by mgraczyk 4y ago
That's basically what this article is about. > “John Carreyrou did a brilliant job of exposing Elizabeth Holmes — I did not.” Auletta reflected. “So, when I look back on that, yeah, I did a profile of her, but I did not produce the
702.
by mgraczyk 4y ago
They are iterative and path dependent, no?
703.
by mgraczyk 4y ago
Lol desktop meaning a physical computer. Engineers still have desks with tops. If anything they have more space than ever since the offices are so empty.
704.
by mgraczyk 4y ago
What? Doesn't even make sense. Why would lacking a physical desktop cause developers to make bad software?
705.
by mgraczyk 4y ago
Interestingly, at Google the typical developer workflow (google3) is very cloud native. Most devs write code in VS code in the browser. Many (most?) devs don't have a physical desktop any more, just a cloud VM. The code lives in a netw
706.
by mgraczyk 4y ago
I've looked a little at this in the context of under display cameras, which introduce similar artifacts. They could remove these artifacts and generate realistic looking images that are physically plausible, but the content in regions
707.
by mgraczyk 4y ago
To first order I think the explanation is just conversation of angular momentum. The moon earth system has to have it's angular momentum constant over time. If the moon is further away in a stable orbit, the earth must rotate slower.
708.
by mgraczyk 4y ago
How could a banner not be sufficient? Under what legal theory is Copilot + banner illegal, but code search is legal?
709.
by mgraczyk 4y ago
You run search on the text output. Yes I mean that they would search the code. It's certainly legal to show LGPL code snippets along with the license (It may also be legal without this, IANAL).
710.
by mgraczyk 4y ago
All this discussion of legality is interesting to me, because I'm pretty sure that if Github ran a search in the background, found the corresponding license for the code snippet, then showed it to the user in some cookie-banner like an
711.
by mgraczyk 4y ago
Interesting, thanks for explaining!
712.
by mgraczyk 4y ago
Interesting, but I guess I'd have to learn APL or J to compare. Without understanding what those symbols mean, it seems much less like "math" than the numpy version. 1 + np.min(np.argwhere(running_sum[1:] == 0), axis=0)
713.
by mgraczyk 4y ago
I think that specific example is very easy walks = np.random.choice([-1,1], (100, 1000)) running_sum = np.cumsum(walks, axis=0) plt.plot(running_sum) Intuitively it seems like the more complicated the problem, the bigger t
714.
by mgraczyk 4y ago
I don't really see the appeal in using a math-like notation for things like this. The J code doesn't look anything like what I would write down to express this mathematically, and the equivalent python code is nearly as short and
715.
by mgraczyk 4y ago
What is the difference between this and describing a human brain the same way? The brain is the model, you are "just" copying things from the memory of your brain to words that you speak or write? I don't think it's peda
716.
by mgraczyk 4y ago
Current LLMs are "modeling" something according to pretty much any sense of the word "model". In the technical, computational linguistics sense, LLMs are language models that give a conditional posterior distribution ove
717.
by mgraczyk 4y ago
It is semantics, but that was your whole point no? > That's because they're not modelling anything If we agree on "how LLMs work", then how can you claim that they aren't modeling anything? They are modeling lang
718.
by mgraczyk 4y ago
I'm disagreeing with the language you are using to characterize models. "copying from memory" implies that there is something being copied, and a memory that you are copying it from. I am pointing out that LLMs do not do this
719.
by mgraczyk 4y ago
This is factually wrong, both in terms of quantity and quality. Current AI models are not "just sort of repeating and copying from memory". This is just an incorrect characterization of how they work and how they perform. AI skept
720.
by mgraczyk 4y ago
They could have said what happens next in just as many characters, but chose not to. I assume the reason is because fewer people would read the article in that case.

More ›