moyix

25 ms·

241.
by moyix 5y ago
A gist of what? The slur list? That's already on my website: https://moyix.net/~moyix/copilot_slurs_rot13.txt
242.
by moyix 5y ago
Yes, they mention this in the Codex paper [1]: > Since Codex is evaluated on natural language prompts, we hypothesized that it would be beneficial to fine-tune from the GPT-3 (Brown et al., 2020) model family, which already contains stro
243.
by moyix 5y ago
Nice trick! Unfortunately at this point the bottleneck really is not on generating candidates but on evaluating them. The existing early-exit code from Sc00bz [1] can generate all the lowercase alpha possibilities up to 13 letters in about
244.
by moyix 5y ago
One thing I do want to clear up - lots of people have been saying things like "Great! So if I want Copilot to stay away from my code I can just include one of these words!" But there are some problems with that idea: 1. The banned
245.
by moyix 5y ago
The way twitter refuses to show you the actual tweet someone linked to in the middle of a thread is super-annoying. One thing that can help a little bit (on the desktop version at least) is to load the URL without the referer (i.e., highlig
246.
by moyix 5y ago
I haven’t dug into others. GPT3 does have some kind of filter (it warns you in the UI if it detects sensitive content) but I don’t know anything about the implementation. Also worth noting that this kind of analysis only really works if the
247.
by moyix 5y ago
Happy to answer questions about the list or techniques used to decode it! The last two words have stubbornly refused to yield so far.
248.
by moyix 5y ago
Yes, unfortunately the code example got somewhat mangled as it passed through the arXiv sanitization script [1]. The original is: char str_a[20], str_b[20], str_c[20]; sprintf(str_a, "%f", a); sprintf(str_b, "%f
249.
by moyix 5y ago
I find demos like yours hard to reason about because it is often the case that you can mess around with the prompt to get it to demonstrate understanding of things that it didn't seem to understand before. I wanted to see if it underst
250.
by moyix 5y ago
I don't know if they have made it into production yet (probably not?) but Lindsey Kuper wrote a nice pair of posts on how DNNs can replace an existing collision avoidance system (ACAS), and how to verify whether it's still safe af
251.
by moyix 5y ago
I have already seen reports from biologists about AlphaFold solving protein structures that have resisted laboratory methods for decades [1,2]. It's hard for me to see why this is not a transformative development for (one part of) basi
252.
by moyix 5y ago
Anyone else getting a 403 Forbidden? If so it might be better to link to the paper instead: https://www.nature.com/articles/s41586-021-03828-1
253.
by moyix 5y ago
I'm quoted in the article linked in the OP, and a couple other recent ones: https://www.wired.com/story/ai-latest-trick-writing-computer... https://cacm.acm.org/news/253785-will-ai-rewrite-cod
254.
by moyix 5y ago
This is pretty much already assumed by program synthesis work AFAIK. There's also work on synthesis from I/O samples (e.g. FlashFill) but you don't get very good guarantees (and sometimes it makes hilarious mistakes: https:&
255.
by moyix 5y ago
I think it's inevitable that program synthesis will adopt deep learning methods, whether or not those look like large language models. There's been a bunch of work on this (that you're probably familiar with) from Dawn Song&#
256.
by moyix 5y ago
Programmers love to pretend that they're lawyers, especially when it comes to copyright law. Something about the law really appeals to hackers!
257.
by moyix 5y ago
What? I don't think I made any claim of the sort. I'm claiming that it does more than mere regurgitation and has done some amount of abstraction, not that it has human-level understanding. As an example, GPT-3 learned some arith
258.
by moyix 5y ago
Have you used Copilot? I have not, but I have trained a GPT2 model on open source projects ( https://doesnotexist.codes/ ). It does not just pattern match and replicate. It can be cajoled into reproducing some memorized sni
259.
by moyix 5y ago
Surely you would have to sue the people using the tool to produce verbatim copies of code, not the creator of the tool?
260.
by moyix 5y ago
It's worth noting that for most papers, arXiv provides the LaTeX source for download, which is presumably what they trained on.
261.
by moyix 5y ago
Bug in turing_awards Q1: input.txt doesn't include the first 4 lines of the file (which are shown in the flavor text). So you have to trick it into giving the right number instead of doing the obvious grep | wc -l :p This also makes Q2
262.
by moyix 5y ago
If you hover over the command it will give a brief summary. The selection of commands and patterns does change slightly for each "level".
263.
by moyix 5y ago
I would love to be able to just type the command line in; dragging blocks around is really slow and unnatural. But probably I'm not the target audience...
264.
by moyix 5y ago
Another helpful option if you're trying to get traces out is `-d nochain`, which turns off translation block chaining (chaining inserts a direct jump from one block to the next, which can cause logging statements to be skipped). Also,
265.
by moyix 6y ago
The author seems to be using OpenBSD, so I imagine that rather limits their options.
266.
by moyix 6y ago
I recently had to just disable the Gmail spam filter using a custom rule [1] because the false positive rate was awful – 90% of the stuff in my spam folder was legitimate, including Google Calendar invites and normal GitHub communication. U
267.
by moyix 6y ago
Yes, to be extra clear, if that is their claim, their claim is wrong.
268.
by moyix 6y ago
To be clear, by "learned to read" – it actually learned to OCR text, unsupervised. That is pretty impressive, IMO. As for whether NNs are doing symbolic manipulation – I'm not so sure that they aren't. DALL-E, the sibl
269.
by moyix 6y ago
Keep in mind that a motivated academic research team was able to write a hard drive firmware implant on their own as well: https://www.impeachdonaldtrump.net/Implementation.and.Implic... But yes, making it work for a bunch
270.
by moyix 6y ago
If you want to look at some more fake samples, here are 256 generated snippets. These ones start at the beginning of the file each time, so you should get a better sense of context: https://moyix.net/~moyix/unconditiona

More ›