thatguysaguy

8 ms·

61.
by thatguysaguy 3y ago
At the end of the day all the arrays are 1 dimensional and thinking of them as 2 dimensional is just an indexing convenience. A matrix multiply is a bunch of vector dot products in a row. Higher tensor contractions can be built out of lower
62.
by thatguysaguy 3y ago
Well, in the transformer forward pass there are a bunch of 4-dimensional arrays being used.
63.
by thatguysaguy 3y ago
Welllll, there seems to at least be some mathematical cheating in that this is representing a non-well-founded set. (Declare each pixel to be the set of subpixels at the next level down defining it, this forms an infinite descending chain).
64.
by thatguysaguy 3y ago
Uninmportant, but if you're citing Moore's paper I feel like you're just trying to pad out the references to make it look like you're serious
65.
by thatguysaguy 3y ago
Too new is definitely one thing. Someone is going to have to make a gamble to actually paying for a serious pretraining run with this architecture before we know how it really stacks up against transformers. There are some papers suggesting
66.
by thatguysaguy 3y ago
In this particular case I've just directly looked at the friend's pokertracker data, which includes rake values and then you just add rakeback to get the total. It's harder to cope oneself in an online setting where there&#x2
67.
by thatguysaguy 3y ago
Lower than software, but about $50/hr
68.
by thatguysaguy 3y ago
I have a friend who's self-studied up to a pretty good hourly rate within ~2 years. The main thing he does is tons and tons of time studying solver outputs.
69.
by thatguysaguy 3y ago
I think it'd be nice to have some harder puzzles, it seems a bit too straightforward once you figure out the tactics ATM.
70.
by thatguysaguy 3y ago
It's not really hidden, that's an intentional feature. All our LMs have been using this (or similar) since GPT-1.
71.
by thatguysaguy 3y ago
I posted a few example questions (which are definitely way below Einstein level) in another comment. It's not doing too well! Do you have any links to conversations where it was particularly helpful? I do wonder if my GPT usage is bad
72.
by thatguysaguy 3y ago
Reverse dictionary is a great one yeah. I think I've used it for resolving a tip-of-my tongue thing several times. For language learning I'm a bit more wary. A few times I've asked it for help with Chinese, and when I ask one
73.
by thatguysaguy 3y ago
Hmm, I tried some questions which would have been relevant to me in the recent past, and it flubbed pretty hard on all of them. Even worse, instead of just saying it doesn't know it generates semi-plausible babbling or adjacent but not
74.
by thatguysaguy 3y ago
Completely agreed on this one, although it's maybe more like 70% for me. I did leave an exception for that specifically in my original comment.
75.
by thatguysaguy 3y ago
Could you tell me some typical uses beyond just asking for code snippets? I'm an NLP researcher, and currently I still find myself using Google for other types of information seeking. I'm working on NLG decoding algorithms current
76.
by thatguysaguy 3y ago
The net flow of knowledge about text-to-image generation from OpenAI has definitely been outward. The early open source methods used CLIP, which OpenAI came up with. Dall-e (1) was also the first demonstration that we could do text to image
77.
by thatguysaguy 3y ago
I'd love to see something like that. The world needs more auctions tbh.
78.
by thatguysaguy 3y ago
Done already, thanks
79.
by thatguysaguy 3y ago
MIT yeah, I should have mentioned. I wasn't trying to say anything illegal was going on.
80.
Ask HN: A better known dev reimplemented my open source repo. What should I do?
17 points by thatguysaguy 3y ago|17 comments
81.
by thatguysaguy 3y ago
What would the right answer be?
82.
by thatguysaguy 3y ago
The open vocabulary aspect seems important. Word2vec would never let me make "Pirate Pope Wars".
83.
by thatguysaguy 3y ago
As an example, look at the second link on HN right now: https://neal.fun/infinite-craft/ I just opened up a discord server I'm in and everyone is spending quite a lot of time on it!
84.
by thatguysaguy 3y ago
I think the problem is pretty interesting though. Better definitions of curiosity might still have this failure mode. Human curiosity definitely does!
85.
by thatguysaguy 3y ago
Try OpenAI's spinning up: https://spinningup.openai.com/en/latest/
86.
by thatguysaguy 3y ago
My personal laptop setup has been arch on a Lenovo Thinkpad for about 7 years now. It's been great!
87.
by thatguysaguy 3y ago
The article definitely doesn't do the work needed to establish that we should care about the headline numbers at all. It states an estimate of how much it would cost to fix things detected by this automated system, but there's no
88.
by thatguysaguy 3y ago
I haven't seen this key/buzzword mentioned yet, so I think part of it is the fact that we're now working on complex systems. This was already true (a social network is a complex system), but now we have the impenetrability of
89.
by thatguysaguy 3y ago
This is definitely one of those things where you don't need a paper to "suggest" it. It's just on the table as a semi-plausible option already. The value in the paper is suggesting concrete mechanisms by which it might a
90.
by thatguysaguy 3y ago
The point about twin studies only showing heritability in reasonably well-off families is one I haven't seen discussed before. That seems like a huge asterisk on all the claims based on twin studies I've seen.

More ›