Ayush Agarwal (आयुष अग्रवाल)

@condret@fedi.absturztau.be if you don't see a problem with LLM boosters effectively inflicting DDoS attacks by ignoring robots.txt, you shouldn't have any issues with those websites serving poiosoned data and malware by detecting you're using LLMs. @mttaggart@infosec.exchange @nohhue@mastodon.social

@jacqueline@chaos.social I’ve been averse to being near people who smoke ever since I was a kid. I’ve met some nice people who turned to be addicted to cigarettes and that always turned me off and I started avoiding them. I can’t stand the smell and realising that I’m passively smoking being around them didn’t help. I’ve been observing the same thing in my feed in the past few months — people who I respected turned out to have “the ends justifies the means” result-oriented brains and I suspect that such brains are susceptible to the temptation of getting addicted to LLMs. It’s sad to see such people fall and taint major projects like Vim.

@behdad @khaled In addition, this comment is just as much a reference for myself as it is a reply. I read the Claude's Cycles PDF written by Knuth. Knuth had already solved the exercise he was working on, himself, and was interested in a generalization of the answer he got for other values. His friend, Filip, gave the question to Claude without bothering to solve it himself? Or maybe he did bother and solved 4 ≤ m ≤ 16 himself, it's not really clear from the PDF. After 30 different explorations, Claude came up short. In the 31st exploration, Claude produced results. Knuth still needed proof and it's not clear if he wrote the proof himself but he doesn't mention using Claude for writing the proof that he did. He then goes on to say Filip told me that the explorations reported above, though ultimately successful, weren’t really smooth. He had to do some restarts when Claude stopped on random errors; then some of the previous search results were lost. After every two or three test programs were run, he had to remind Claude again and again that it was supposed to document its progress carefully. So I don't see how that PDF leads to the conclusion that "if it's good enough for Knuth, it's good enough for me". He's impressed with the results, sure, but it seems like he didn't even use Claude himself? His friends did and they had to repeatedly nudge it to steer it in the correct direction. Regardless of what I said in this comment and the results of the exploration from Knuth's friends, appeal to authority leaves a bad impression.

@TheOneDoc I feel it's not as if most people can't code, they can do it just fine. It's just that most people, in the pursuit of writing code and earning money, don't assign any thought or importance to things like the politics behind free software, the issues surrounding copyright laws, and how code gives an individual immense power and freedom that can rival the impact a corporation may have. One of my major sources of inspiration for choosing computer science and being interested in free software was the documentary TPB AFK and The Internet's Own Boy: The Story of Aaron Swartz, not money or lines of code or cool software features or "blazing fast" performance or memory safety. One of the (flawed) arguments I've read from people who cheer for LLMs is that they "democratize" access to knowledge and "liberate" it from the shackles of copyright. It's unsurprising to see that people don't realise or care where these LLMs get their data from (by DDoSing websites) but it is surprising to see them not realise who ultimately controls these models and software and the training data behind it. Claude can't "democratize" knowledge because it's limited by the amount of tokens one has access to and the same goes for all the other LLMs. The access may be generous and subsidized at present because of a desire to make people dependent on it but it won't be like that way forever. I mean, we have seen this pattern before with streaming platforms as well but somehow the temptation in this case is too strong to ignore the wisdom people may have gained in the past? It's baffling. The argument that LLMs liberate knowledge from copyright restrictions is hilarious. Apparently, the actions of corporations and firms who lobby for copyright and DRM isn't enough to make people realise that copyright can always be used by those in power to oppress individuals when it suits them, even at present. I mean, good luck defending oneself in court if a corporation decides that your LLM generated software infringes on their copyright. However, individuals never had and never will wield such power. All LLMs do is pit individuals against each other and give corporations even more power to violate copyright when it suits them and harass people when it doesn't. @raulinbonn @liw

@atax1a I mean, I fail to see how it wasn't obvious to anyone who has a shred of curiosity and understanding about how copyright works. The fact that we were able to predict it doesn't really feel all that good to me and I don't feel like saying "I told you so" to people who gleefully argued that the existence of LLMs means "knowledge is liberated" or "copyright is dead" or "individuals can reverse engineer proprietary code using LLMs". Yeah, good luck with that. @ariadne

@Joshsharp if I had to guess, there are no obvious consequences of doing so. One doesn't need to worry after DDoSing the websites of multiple people on the Internet, not unless one does it to a powerful and wealthy person or corporation.

@arun@fantastic.earth I'm not sure if not knowing how to list files and directories in a directory in a list format in reverse chronological order with directories grouped together and the trouble of figuring it out by reading an esoteric man page would be considered a UX failure. I've seen some sysadmins who don't even bother looking at man pages or using the help flag, before and (especially) after LLMs are a thing. Of course, not all man pages are informative or well written but figuring things out, and hopefully improving them, is part of learning. With LLMs, we jump straight to the result and never learn or improve anything. @jschauma@mstdn.social

@arun@fantastic.earth I did accept that not all man pages are well written and in those cases, they are a UX failure even for an experienced sysadmin. I also accept that something I find easy to do at this point (using sed, for example) will be considered arcane by some people. However, not everything can be easy to figure out in life, regardless of how excellent our documentation and onboarding is. Learning stuff takes time. Even using a bare bones GUI file manager like Nautilus will take time for someone who has just migrated from Windows. If I try to learn how to use Racket or Scheme to replace my CommonMark files, it'll take time. I can throw an LLM at it to do that for me and it'll take seconds to do something I might take a few months to do but there will be no learning or scope of improvement or creativity in that case. I guess it depends on the context? I don't have to understand how Wayland works to use a compositor like Sway. An abstraction like that is practical and expected. However, one has to question if such an abstraction is really healthy in cases where learning and improvement is the objective. What if we never teach kids how to manually compute the square root of a number or never tell them how to figure out the hypotenuse of a triangle and just ask them to rote learn the formula and use a calculator? @jschauma@mstdn.social

@Codeberg@social.anoxinon.de I'm happy to see such a stance but it'd be nicer if there was some decision about what I proposed a few weeks ago.