Apparently theres an AI that can make simple software now

How fuarrrked are we? https://github.com/OpenBMB/ChatDev Muh body is empty!!!

>>37045 Not fuarrrked at all. Many people are far better than me in my fields I'm interested in(both artistic and non-artistic field), it however never stopped me from enjoying doing those things. No matter how good I'm at playing Piano, Chopin will always be far superior to me. If a computer comes along and is also better than me, perhaps even better than Chopin, it doesn't change anything for me at all. I do things because I enjoy doing them, not because some machine or whatever is unable to do it.

Waow! A new dev tool! but nah it's really just novelty for now it seems. Cool project tho. Would love to see an AI that is trained to spot vulns / CVEs and bugs though. Would be much more useful than this. I'm not too bothered about AI overtaking software dev. If it ever will.

completely fuarrrked muggles [and other retards like lainniggers] believe that random idiotic practices like string concatenation are valid. they literally still have security footguns in XML in 2023. everything is made by droolers. AI is trained on that. expect new classes of bugs and vulnerabilities that did not exist previously. you will press the button to run your laundry machine. instead it will run your dryer

>>37045 None of this neural network nonsense will replace intelligent people anytime soon. >>37161 >muggles [and other retards like lainniggers] believe that random idiotic practices like string concatenation are valid. I don't, but I'm also not mentioning it in every thread. Go post in the thread I mentioned already. >expect new classes of bugs and vulnerabilities that did not exist previously. This is a correct observation. The only proper way to have machines generate programs is through correct methods, through mathematical manipulations, which is what a compiler is, and the compiler doesn't remove the need to tell it what to do properly.

>OpenAI API Key GPT4 is good at coding, news at 12. Wake me up when I can run something like this locally with an actually open AI. I've tried codellama and codegeex, and so far I'm disappointed in the results. >>37048 Healthy view to have. It's not good to base your self-worth off comparisons to others. Especially when the others are machines. >>37165 >None of this neural network nonsense will replace intelligent people anytime soon Unfortunately most current work doesn't require much intelligence. I think we're going to see a lot of jobs disappear and a revival of ludditism. It'll hit doubly hard if there isn't a revival of cheap capital to pay bullshit jobs. Already I see a lot of outsourcing, and once AI can be more cost effective (not the same as better) than that, the genie isn't going back in the bottle.

>>37168 >Unfortunately most current work doesn't require much intelligence. Yes, that's what I meant to imply. Regardless of what happens, we live in interesting times.

>>37168 > and once AI can be more cost effective I feel like this is still an "if" rather than a "when".

>>37168 Outsourcing doesn't really work the way most people assume it does. The legion of sirs is just for the filtering screen, but a valuable contractor you can rely on when things go south is worth gold that most companies pay at silver rates. The thing with outsourcing is that it's a cost-saving measure, not an easy way to get a sweatshop going. Like all other corporate tricks it has its ugly downsides though.

>>37168 luddites and sabotage are both terms that come out of the industrial revolution, an event which lead to the rise of significantly more formal employment at the same time as destroying historic trades and causing the rage that these words signify. You should expect this rage whenever someones livelyhood is taken away, or lifestyle rendered impractical. But people seem to move on, or at least their children do.

>>37045 I'm debating making a new thread to discuss AI/LLMs but for now I'll take over this thread since it kinda relates; don't want to kill a thread yet, since I know this board is slow. I know a lot of people keep saying conflicting things across the net about AI/LLM. The worst is when I see dubious articles about people using ChatGpt to make marketing campaigns or coming up with product ideas, coding said products, ect. They never mention what version of ChatGpt they use and never post the results. I know AI/LLMs is going to change the landscape of tech, but so much of it feels like fizz buzz or exaggerated to the point it feels like its feeding a .COM bubble. Not to mention to my knowledge OpenAI hasn't disclosed their databases on how they got all this data to train on. That a lot of it is scraped data or using data and basically reselling it as a new product. I personally think they should have to disclose all their data and I'm glad to see some famous people and creators suing them for using their content without permission. What are some of your thoughts? Where do you like to go to get updates about AI/LLM? It seems more people in the industry, at least coding, don't see it as impeding doom, depending on their skill level. I definitely see it being useful on working with huge code structures; I know IBM wants to use AI to detangle their Cobol code bases. I really don't like AI """art""" and hope to see copywrite and laws adjust to make it not replace artists. Though using your own data set and training it to make new images for you sounds interesting. Also AI for use in making memes and for game moding awesome. P.S. Prompting Engineer will always sound like a bullshit job title.

>>37966 >I know AI/LLMs is going to change the landscape of tech, but so much of it feels like fizz buzz or exaggerated to the point it feels like its feeding a .COM bubble. Very much this. There is a lot of cool soykaf being done in the machine learning world and with LLMs but you have to put a heavy filter on everything to get to the meat. Every field of academia and industry has an incentive to say that AI is an amazing new technology that is transforming everything, because that is where the money is now. GPT and stable diffusion are both amazing pieces of software but they're more like Cleverbot than AI. >Where do you like to go to get updates about AI/LLM? I don't, I wait for friends to link me papers they find interesting. Scott Aaronson is reliably knowledgeable and honest, he's also a great source to go to whenever quantum computing is in the news. >I definitely see it being useful on working with huge code structures; I know IBM wants to use AI to detangle their Cobol code bases. Yes, a friend of mine has also told me that Intel is starting to use LLM's internally to make it easier to find documentation.

>>37168 >>37168 >Unfortunately most current work doesn't require much intelligence. I think we're going to see a lot of jobs disappear and a revival of ludditism. It'll hit doubly hard if there isn't a revival of cheap capital to pay bullshit jobs. Already I see a lot of outsourcing, and once AI can be more cost effective (not the same as better) than that, the genie isn't going back in the bottle. I think you hit the nail on the head without realizing it here. AI will replace outsourcing. Outsourcing has not completely gobbled software positions yet. Implementing from a spec is the least effective way to spend my time anyway.

>>37966 I can imagine a programmer some 30 years ago lamenting if he knew programming would be pure lego-blocking with little regard for design. Well, mainstream programming. I figure a next logical step would be to teach the thing to produce correct working code, and even to run the tests themselves. Comparing how our present day "software engineers" understand the code they are working with, there would be little difference. Indeed, the replacable programmer, the cogwheel will always be replaceable. We would be a lot better off without them, but as thou hast pointed out, they would flock in teams to try to steer the thing. I bet you we'll be seeing projects written the bulk of by gpt. There are places where they use golems to do stuff like manually pumping water, but here we still use flesh. human rites n stuff. I imagine a version of the language model that can be told "define a function with a main loop, with such and such variables...", "test it with random input, including faulty data. I tried this with gpt and didn't work. Gpt is fuarrrking dumb, btw. Like if it can take your job, a pulley probably can just as well.

>>38176 >Like if it can take your job, a pulley probably can just as well. Lmao. It's true. The fact that so many technically proficient people think gpt can teach them something baffles me.

>>37966 I think it's really cool and most of the negativity is Twitter-originating groupthink. Copyright is like kicking dead whales down the beach and does not need to be expanded, especially not in a way that will in practice render corporations the only ones legally allowed to use AI (because they outright own large enough data sets). There is nothing unethical about using copyrighted training data just like there is nothing unethical about media piracy. I will continue using artist's names in my prompts, and I will continue to use LLMs as a coding assistant, because AI is objectively very powerful and useful. fuarrrk the people trying to get local models regulated and locked down.

>>38179 Would databases of rather specialized data really help corporations enhance their own LLMs for more general purposes? Like I can run a local instance of mixtral right now- would it be much better if I trained a model up based on all the residual data sticking around a startup incubator or major consultant firm? I don't know what kind of data they have but I would think legal documents, emails and powerpoint presentations contribute little unique content. If I were an academic publisher and trained it up a bunch on papers I can imagine that being very neat, possibly very insightful about topics - but would it be much better than just using grounding techniques to treat it like a database?

>>38184 I am thinking more along the lines of, if using AI created without copyright over the training data is deemed illegal, as dumb angry artists seem to want, then you won't be allowed to openly run a local instance of something like mixtral, because mixtral was trained without regard for copyright. In this scenario I think what would then happen could be along the lines of, a tech company strikes a deal with Reddit (a major source of training data, which has recently hobbled itself to prevent scraping in part for this specific reason) to be allowed to legally use their data to make a copyright kosher AI (presumably more powerful than public domain trained alternatives), and then only offer paid access through an API and not local models because Reddit insists on perpetual royalties. For image gen replace Reddit with Disney or something. Don't really think this is actually likely, but it's why the clamoring for 'ethically trained AI' with the idea that they're defying the power of corporations somehow is very stupid.

>>38179 >>38185 Says the man who cares what nobodies on Twitter think; enough that you're being a contrarian out of spite. You can think whatever you want of the situation (I honestly don't have a strong opinion either way), but I don't think it's unreasonable for artists to feel uncomfortable with a program being fed their product without their permission, especially when people like you then use said technology to try and obsolete them. At the end of the day they're human beings, not just living content machines, and most human beings don't like the idea of their jobs being taken away.

>>38186 I don't think that discomfort is unreasonable either, but the proscribed solution definitely is, and a style based notion of intellectual property is especially unreasonable. My preferred solution would be for society to pay artists unconditionally, regardless of the technology. Is your only real objection to my tone here? I don't think it's unreasonable to be a little pissed off about the state of discussion on this topic online for someone who likes AI and thinks it is a hope for positive transformation. I do care what nobodies think, and I wish they would put more consideration and less conformist signalling into it.

>>37357 this isn't as true as it was 10-15 years ago. Lots of contractors in Europe, LatAm and even SEA are getting US contracts because they offer a level of quality that sirs don't provide, for a price that American devs would never accept.

>>37168 Try the latest, largest deepseek. Be surprised.

>>38297 It didn't understand my question when I asked it to write a program with worse than O(2^n) time complexity. Its knowledge of mathematics would also be considered quite superficial if it were a person. Mediocre software.

>>38185 >In this scenario I think what would then happen could be along the lines of, a tech company strikes a deal with Reddit (a major source of training data, which has recently hobbled itself to prevent scraping in part for this specific reason) to be allowed to legally use their data to make a copyright kosher AI (presumably more powerful than public domain trained alternatives), and then only offer paid access through an API and not local models because Reddit insists on perpetual royalties. For image gen replace Reddit with Disney or something. dumb question, why can't someone just scrape old.reddit.com like a good old webpage? they'll have to pass up on "NSFW" posts' contents but still

>>38806 It takes more effort and you have to know how to have it automatically switch VPN / data center IP addresses (ideally IPv6, though Reddit currently does not support IPv6) and know how to get them in bulk, to handle when each individual address gets rate limited by the site. Russians, Chinese, Israelis and North Koreans know how to do this sh!t like the back of their hand, but Western people, including FOSStards (Richard Stallman wrote me a long e-mail telling me how piracy and scraping is immoral and should be strongly condemned), are too scared to think outside the box, to do anything differently from what their university professors told them to do, to risk doing anything "unethical" and to upload anything that would "violate the guidelines" of their precious Microsoft-owned DMCA-obeying GitHub® repository (you're not a real programmer by the "God Bless America" normee IT industry and academia if you don't chase after your precious GitHub® stars). Reddit Inc. is a multi-billion dollar publicly-traded corporation with CIA cells embedded in the company but FOSStards and academics will tell you it's immoral to hack or scrape their data (while the 1% ownership class laugh themselves to the banks with their yachts, drugs and underage hookers).

>>38809 based answer, thank you

>>38809 Unrelated, but I'm using a RateMyProfessors API to do a simple app for my fellow college students. Would it be way more impressive on my resume to learn to scrape it or does it depend on what job I'm going for?

>>38806 >they'll have to pass up on "NSFW" posts' contents but still just give the tool the cookies or use headless browser

>>39738 I'm not sure which one would be more impressive, but both should be easy enough. For the scraping, parsing HTML with regex is a funny meme but not actually a super great idea if you value your time. Look into "html-xml-utils", particularly hxextract and hxselect (you may sometimes need to run the page source through hxclean/hxnormalize first). The man pages for those tools are kind of gay and unhelpful so just post if you can't figure it out and I'll try to remember what I did. Use a browser extension that exports cookies for a specific website in Netscape format and check curl/wget man page to see how to supply it with the cookies.

>>39746 I'll take a look into this, thanks. Might be a while, I'm finishing up school and starting a front-end internship. Do you have a discord or something? There is also RSelenium method and a Python/NodeJS/Puppeteer method. I'm looking into all of them.