Posts by gray17@mastodon.social
(DIR) Post #ATmncDO3qNHwjhs1RI by gray17@mastodon.social
2023-03-19T21:35:45Z
0 likes, 0 repeats
@TedUnderwood @dh This is an interesting application, and I expect more things like this will be useful in the future. But right now, I'm still somewhat skeptical that when you ask GPT for an "explanation of its thoughts", the explanation has any connection to reality. It seems likely to be as much a confabulation/hallucination as anything else, which can be useful, but I don't see how to have confidence that it isn't just hard-to-detect nonsense.
(DIR) Post #ATms2g3uFlN112Ji4m by gray17@mastodon.social
2023-03-19T22:25:22Z
0 likes, 0 repeats
@TedUnderwood @dh Right, but that works only because of statistical patterns. The "explanation" is derived from the previous text, but it's not clear to me that it won't be misled by patterns that it created itself. It's very easy to fool GPT with eg, a riddle that looks like something it's seen before, but has a small difference that makes the answer completely different.
(DIR) Post #ATnHGknXwGwYc3TQIq by gray17@mastodon.social
2023-03-20T03:08:02Z
0 likes, 0 repeats
@TedUnderwood Sure, when you know the answer, you can tell that the generated text is wrong. What bothers me about these models is that the generated text is almost always plausible, so that if I don't check carefully, I might not notice that it's lying about what it said earlier. "Give an answer and explain your reasoning", sometimes it's obvious if the explanation is about an answer different from what it gave. Sometimes the error is more subtle, and I don't know the implications of that
(DIR) Post #AUgtCALSkoDkh8xLl2 by gray17@mastodon.social
2023-04-12T12:29:13Z
0 likes, 0 repeats
@anttipeltola @ibzan The 70s was when the idea of "maximizing shareholder value" was invented. I had heard of the concept as a kid, but I didn't learn until much later that the idea is a fairly recent invention.The idea basically turns corporations into paperclip maximizers. Value-maximizing corporations are the poorly-aligned AI that futurists are worried about, but they're already here and already slowly destroying us.
(DIR) Post #AUjvfiL7swBzwicJAu by gray17@mastodon.social
2023-04-17T10:13:57Z
0 likes, 0 repeats
@mitchmarq42xyz I've learned I'm usually happiest if I throw out what I did and start over. Even if I don't copy anything from the original, the second time is much faster, since I now have a better idea what I'm doing, and it's easier to modify and maintain too.Of course this is disruptive and takes a nontrivial amount of time, so I don't apply this wholescale to anything really large or collaborative. For those, I try to do big rewrites incrementally when feasible
(DIR) Post #AXBg44GTU73147D0FM by gray17@mastodon.social
2023-04-29T03:29:47Z
1 likes, 0 repeats
about #tldr:Sometimes I see an article that I want to show to a friend, but it has a lot of "fluff": intro/background, dramatic narrative, supporting detail, etc. I'm not against fluff, but I summarize, so the friend can decide if they want to poke through the fluff.I thought I'd do some here too. Making legible summaries in 500 chars is an interesting writing exercise.I'm using prefix-less URLs that aren't links, because mastodon adds preview images that are usually irrelevant noise.
(DIR) Post #AYXcI9Tyhx2TfQZnIu by gray17@mastodon.social
2023-08-09T03:08:41Z
0 likes, 0 repeats
@mitchmarq42xyz that's my favorite method of cleaning complex configs: start fresh and bring over anything I miss. It's basically copying GC. I'd do it in real life too if I had a spare house I could move to before burning down the old one
(DIR) Post #AZIPIb0ktY7rDB1QO0 by gray17@mastodon.social
2023-08-31T15:30:21Z
1 likes, 0 repeats
@zorinlynx my favorite depiction of yak shaving:https://www.youtube.com/watch?v=AbSehcT19u0
(DIR) Post #AbQTxJYucZukOek1A0 by gray17@mastodon.social
2023-11-03T10:06:44Z
0 likes, 0 repeats
@simon @MattHodges I feel like "signal for level of information" is maybe misleading. (the paper's claim of "understanding emotion" is definitely misleading).Maybe, "nudge the probability space of plausible continuations"? A French prompt will nudge the plausible continuations to be French, etc. In general, any contextual cues in a prompt will nudge the plausible continuations to make them better fit those contexts.
(DIR) Post #AcOq10zciMzi4lSJhw by gray17@mastodon.social
2023-12-02T12:56:22Z
0 likes, 0 repeats
@simon this seems like a bad description to me? It's the wrong hand, the desk isn't cluttered, there aren't any filing organizers, there aren't walls visible, the suggestion of "a collector, a bookstore owner, an academic, or an artist" is a weird bit of stereotyping
(DIR) Post #AcYLUDowwhczGCVe5Y by gray17@mastodon.social
2023-12-07T03:01:34Z
0 likes, 0 repeats
@simon this is fascinating, and it seems to me similar to how chain-of-reasoning prompting helps LLMs.my mental model of LLMs is still basically "spicy autocomplete": there are a billion mindless agents, each with their own proposal of what to say next.normally, the dominant agents are majority concepts, and any agents "aware" of the unusual sentence are drowned out.but adding the context "most relevant sentence" changes who wins the plausible-completion battle.
(DIR) Post #AcsGuA2R1dHOXO7fFI by gray17@mastodon.social
2022-12-09T21:22:43Z
0 likes, 0 repeats
Pets - at Further Confusion 2020 - 📷 me - #FursuitFridayThis is my second attempt at posting. First time, I added a long alt text, and then accidentally deleted the image, so now I have to write it again.
(DIR) Post #AhZd2Gx2LzNc2FX69I by gray17@mastodon.social
2024-05-05T07:15:17Z
0 likes, 0 repeats
@john Hi, your website https://johnconway.art/ is giving me a 502 error.
(DIR) Post #AhrMqeFXns44Oju4SO by gray17@mastodon.social
2024-05-13T20:38:48Z
0 likes, 0 repeats
@0xC01DC0FFEE nicky case has done a lot of cool stuff.I also like the one on voting https://ncase.me/ballot/and the one on diversity https://ncase.me/polygons/
(DIR) Post #Am4iAK9UCUHMMZw7Y8 by gray17@mastodon.social
2024-09-16T22:37:17Z
0 likes, 0 repeats
@tomjennings @ArchiteuthisFlux that's a syntax error. you can't start an expression with `[`you can do `2[a][3]` or `2[3[a]]`(godbolt.org is useful for quick C experiments)
(DIR) Post #ArzVytnChbhBMel95U by gray17@mastodon.social
2025-03-12T19:28:32Z
0 likes, 0 repeats
@wollman @b0rk my vague memory is that the context switch cost was significant. this was born in the era of machines with 64KB of RAM and CPUs slower than 1 MIPS, supporting maybe 2-4 simultaneous users on teletypes and early CRTs. the tty driver also handled serial interface aspects like baud, cr/lf delays, etc.
(DIR) Post #AwZ45kpPDiYZmxbFMe by gray17@mastodon.social
2025-07-27T16:45:04Z
0 likes, 0 repeats
@davidrevoy the sailors end up fighting each other, "she's mine!" etc, wrecking themselves, while the mermaid looks on with annoyance, "not again"
(DIR) Post #AySR0m2ycvlunGWcGe by gray17@mastodon.social
2025-09-22T07:43:22Z
0 likes, 0 repeats
@BrodieOnLinux see alsohttps://pixieditor.net/blog/2025/07/30/20-release/
(DIR) Post #B1D6MzEQOGJyLdAkPw by gray17@mastodon.social
2025-12-13T15:47:40Z
0 likes, 0 repeats
@futurebird this is the one thing I'd do with watch-my-life AI: tell it to nudge me away from any product that's been advertised to me
(DIR) Post #B2B5xOJ98f4tb2qITI by gray17@mastodon.social
2026-01-11T14:25:43Z
0 likes, 0 repeats
@futurebird I've never taught a class, but my first thought is to try starting with a live-coding session where I make mistakes myself