joshka

38 ms·

541.
by joshka 3y ago
It's more that those tools were what come to mind when I specifically think of my exposure to the existence of rust. Its perhaps not that the tools were there, but that they were well known (and known for being written in rust). Anecda
542.
by joshka 3y ago
In answering this, I'd look beyond just the language features and look at the background of how it caught on. What apps were built first. When did the "Rewrite it in rust" mantra take hold. What impact did <system tool>
543.
by joshka 3y ago
https://github.com/bigskysoftware/dotmeta/issues/4 --- While probably a bikeshed consideration, .config is a much more obvious name to use for this. HN thread with many agreements: https://new
544.
by joshka 3y ago
Take a look at youtube vids for this. Mainly because you're going to see people show all the steps when presenting instead of skipping them when talking about what they did. E.g. https://www.youtube.com/watch?v=KenORQDC
545.
by joshka 3y ago
It's much better if you actually provide the context. Here's a better prompt that's mostly just a copy of the advent of code. Only the first two sentences are changed (mainly because the problem doesn't mention anything
546.
by joshka 3y ago
Thanks! I didn't know that. :D
547.
by joshka 3y ago
The immediately obvious question is what happens if you write the comments as doc strings in the correct tense? “Reads” rather than “read”. I wouldn’t be surprised to see that make a meaningful difference.
548.
by joshka 3y ago
Those are bad prompts. Not sure if you're being intentionally ironic here in demonstrating the parent's claim. --- Revise the following paragraph that serves as a reply to a comment on Hacker News about people's per
549.
by joshka 3y ago
Thanks!. The canonical url is much better than the one I found. I've seen it mentioned that this originally came from Usenet groups. I'm pretty sure it wasn't created in this final form, as I recall seeing it evolve in the ea
550.
by joshka 3y ago
Does anyone remember the Slashdot Spam Solutions Checklist?[1] Here's my effort to try to build an LLM is Bad at <TASK> Checklist (: --- This critique of LLMs for writing software is (x) ill-informed ( ) outdated ( ) b
551.
by joshka 3y ago
Try https://github.com/marionebl/svg-term-cli for asciinema -> moving svg
552.
by joshka 3y ago
TBH, I think the answer to this is to fill the knowledge gap. Exactly how is the difficult part. How do you make "The moon is made of rock" be more likely than "The moon is made of cheese" when there is significantly mor
553.
by joshka 3y ago
There's a good slide I saw in Andrej Karpathy's talk[1] at build the other day. It's from a paper talking about training for InstructGPT[2]. Direct link to the figure[3]. The main instruction for people doing the task is: &qu
554.
by joshka 3y ago
To some extent it does contain facts, but those facts are indistinguishable from non-facts, are just a small proportion of the dataset compared to everything else are indistinguishable from non factual information. E.g. using text-davinci-0
555.
by joshka 3y ago
@dang - the earlier article on reason.com is more thorough - perhaps merge these two posts? https://news.ycombinator.com/item?id=36092509 https://reason.com/volokh/2023/05/27/a-lawyers-fi
556.
by joshka 3y ago
> Copyright violation is generally on the output side of things, not the input. It's not a copyright violation to watch a movie. Copyright infringement is doing anything that violates a exclusive right of a copyright holder[1]. Ther
557.
by joshka 3y ago
Interestingly a 2 days prior to https://arxiv.org/abs/2305.10601 , someone released https://arxiv.org/abs/2305.08291 > Large Language Model Guided Tree-of-Thought > In this paper, we introduc
558.
by joshka 3y ago
Not sure if it's relevant, but the OpenAI APIs generally support taking multiple responses in a single API call. I'm unsure what the generalized effect on processing time of that is however. From what I've read, this is sub-l
559.
by joshka 3y ago
Including the standard Slashdot line: "I for one welcome our XXXX overlords" in the training corpus was a mistake we should have seen coming.
560.
by joshka 3y ago
> Copyright infringement is when you take copyrighted work and distribute it directly, or so close to directly that it can't be said to be "transformative". Distribution isn't necessarily a required element of infring
561.
by joshka 3y ago
This is a bad take that doesn't account for the legal definition of copyright. The first step of training an LLM, going from the source to a tokenized input array is creating a derivative work. Unless that's something that can be
562.
by joshka 3y ago
Yeah, I'm also curious about a) round trips and b) how much would have to be doubled (is there a new endpoint that keeps the existing context while adding or streams to the api rather than just from it?)
563.
by joshka 3y ago
As an incidental observation, technical mailing lists could be a goldmine for training LLMs about software engineering practice in the real world. I wonder if these end up in any of the common corpuses used for model training.
564.
by joshka 3y ago
You missed a chance to link to the trace buster buster buster scene from the big hit: https://www.youtube.com/watch?v=2VY_xxL2jL0 (language warning)
565.
by joshka 3y ago
This is 525 tokens, so it's reasonable, but not blocking on the larger models.
566.
by joshka 3y ago
You could run tests against how well each rule is implemented.
567.
by joshka 4y ago
The exact full prompt and methodology is missing from the paper[1] and Github repo[2], which makes this non-reproducible and not particularly useful in understanding the actual issue. I really think this sort of research is critical to do m
568.
by joshka 4y ago
It seems weird to me that you'd want a model per API to be trained rather than a model that understands many APIs. There's likely more similarities than differences in that the differences are really your domain model which infect
569.
by joshka 4y ago
This is just a state of play issue. It's not good enough YET . On cost: 3.5 is 1/15th the cost, and a bunch faster though supports less context, so it's worth experimenting with which parts of tasks need GPT-4 and which need
570.
by joshka 4y ago
My problem is that in communicating that LLMs can’t be trusted by stating that they are lying we introduce a partial falsehood of our own. I agree to some extent that we need a snappy way to communicate this (and that the proposed wording I

More ›