Show HN: Let's Build AI
letsbuild.aiHi,
I built letsbuild.ai over the weekend. This is a link site dedicated to AI and can be edited via Github.
There is well over 50 links already as a start. If you got interesting AI related links/ categories, would welcome any PR. I'm currrently re-reading Jurassic Park, and I can't help but wonder whether we are committing the same mistakes with AI as the scientists in the book... AI seems to be the ultimate example of "shoot first, ask questions later". We have no idea how AI will effect society, the marginal benefits to people are near-zero (measred in terms of the difference before any adoption to widespread adoption, not incremental intermediate stages), and most of the effects will likely be disastrous (much more efficient propaganda, much more efficient drug-in-the-form-of-media production to distract us from the true ills of the world to prevent a true revolution for good, and a huge wealth concentration towards tech companies). This wanton development of AI seems exceptionally reckless... > AI seems to be the ultimate example of "shoot first, ask questions later". We have no idea how AI will effect society You know, you say this, but we had massive changes after personal computers and the Internet and the iPhone became a thing, and no studies to my knowledge were produced before any of these things were introduced. All of the predicted effects were only positive, because only good actors were involved in creating them, but as we now know, a lot of downsides also resulted, mainly because of bad actors and perverse incentives and simple human psychology. The progress of the last 40 years seems to be a validation of "Be careful what you wish for" I used to bike 5 miles to a library in the 80's (as a teen) in order to learn anything. Now it's all on a device in my pocket that I might have killed for were it available then. But of course, it's not just the phone, it's the ubiquitous cell network, the (wired) Internet, display/memory/cpu tech and everything else that all makes it possible. Can I spend "too much time" on my phone? Sure, but I'm also hyper-aware of that tendency now. I've learned to cope with the downsides, while gaining the advantages. I tend to believe that tech becomes available when we're ready for it, and that the positives always outweigh the negatives. (I'm still on the fence here regarding arms.) The obvious difference, though, is that you knowingly use all of these tools; you get to choose whether and how you use them. AI, like the weapons you point out, can invade your life without you knowing it even exists. It's actually the other way around with AI. If you want to do web-search, you got to disclose your keywords to a search engine. If you want to post on social networks, you are in plain sight. If you use a mobile phone, you broadcast your location. But if you use your local models, it's more like running a linux box, it is private, customizable and unfiltered. LLMs will reverse the centralization trend we have seen with search and social. They can create a "safe space", "a room of one's own" where we can be creative and unrestricted. LLMs promise the privacy internet never gave us, anyone, remember how we felt private online 20 years ago? You can't download a Google or FB but you can download a Mistral and even run it on a normal laptop. > But if you use your local models, it's more like running a linux box, it is private, customizable and unfiltered. LLMs will reverse the centralization trend we have seen with search and social. They can create a "safe space", "a room of one's own" where we can be creative and unrestricted. i really like this idea too. Running LLMs locally reminds me of running Linux locally, it wasn't full on UNIX but it as close to a real operating system regular people could get. Maybe the LLMs you can run at home aren't the bleeding edge but they're as close as you can get. Notice how the technology companies are already lobbying authorities hard to try and keep LLMs only under their control. ..hosted in their 'cloud' and accessible only by paying their subscription. The generation that created the OSS movement and the best success stories, GNU and Linux, are doing their best to make sure it never happens again. edit: fixed some bad grammar The amount of people that will run their own models is negligible at the scale of the society. You also can't choose to cut yourself off from the bad outcomes, no matter how much you try. Propaganda 3.0 might not affect you directly, but the political outcomes it produces will. > The amount of people that will run their own models is negligible at the scale of the society Let me transpose your sentence back to 1980: "The amount of people that will run their own computers is negligible at the scale of the society" (I was 8 in 1980. People literally said this.) All someone needs to do is to productize a whole-house LLM that does voice to text, runs the LLM, does text back to voice (possibly mimicking whatever voice you want), and you'd have an Alexa replacement that is far smarter. I'd buy it in a heartbeat just so that I wouldn't have to do maintenance on it. But people don't run their own computers anymore. A phone on their own is nothing without email maps messenger apps browser cloud storage etc... So yes, nobody runs their own computer anymore. It'll be the same for llms. Can they run on commodity hardware? Yes. Will people go to lengths (including buying better hardware) to run gpt-lite instead of the much better internet-aware cloud llm gpt8? It will become mainstream in a few years when every laptop, cellphone, web browser or operating system will sport local LLMs. For now it is a bit hard, but only a bit. The same AI that invades my life to do harm can "invade" my life and be (extremely) helpful. And the latter is more likely. Because the willpower of an AI is just the mimicked willpower of the people who made it, and there will always be more good people than bad people. You could say the same thing about computers, but you are comfortably sitting in front of one of them. Previously you were probably sitting in a public transportation vehicle. How do you personally distinguish between "this is an apocalyptic mistake" and "this is a step in the right direction" ? > How do you personally distinguish between "this is an apocalyptic mistake" and "this is a step in the right direction" ? Good question. The first step is to analyze each technology individually, rather than usual analogies :) Exactly! And study how it's actually is, without resorting to works of fiction made specificallly to entertain and be dramatic Not cp but I would say it's the scale and timeframes I came to post the same. Computers and the Internet evolved over decades before what we would recognize as modern computers. Their first use cases were mathematical, scientific, and business oriented (word processing, financial models, etc.). It took decades for mass adoption and wider use cases to evolve. Sure, there were many cracks that had to fixed along the way (malware, botnets, privacy, piracy, distribution channel for illicit materials, environmental cost, etc.), but none of those things manifest in a timeline remotely close to what we see with gen AI seemingly reaching widespread adoption in a short ~24 mo. time period. The advancement has been so fast, that the real-world guardrails haven't had any time to catch up before the next advancement completely changes the playing field. Prime example is that Will Smith spaghetti video to Sora within months. Now we have AI denying care[0] to patients and generating deepfake porn[1] of celebrities, coworkers, and even minors. What are the lines we should draw as a society? It moves so fast that even knowing what the right discussion context is can be a challenge because the endpoints have shifted by the time we even understand the problem enough to have a constructive conversation. The socioeconomic impacts in some industries where AI can replaces workers will be tremendously negative for real-world, flesh-and-blood human beings who may not be ready or in a position to adapt to those changes. There is some postulation that the Arab Spring arose out of the effects of climate change on crop failures[3]. How will such a sudden and massive shift in employment and household income combined with other macroeconomic factors like inflation and housing costs affect the stability of society and social unrest? Don't get me wrong: this is not an argument against AI because certainly, if AI can do those jobs better, faster, and more efficiently, then why not? But rather as a community and society, we also need to understand the wider consequences of wiping out employment of 90% of the people in a given field of industry and what that looks like for our friends, families, neighbors. [0] https://arstechnica.com/health/2023/11/ai-with-90-error-rate... [1] https://www.euronews.com/next/2024/03/04/deepfake-porn-video... Unfortunately another theme of Michael Crichton books is that because of human hubris and corporate greed, the disaster is inevitable. Regulation and scientific ethics don't matter because humans find a way. Although, maybe that's because it wouldn't make for a very interesting book if government regulation successfully prevented Hammond from experimenting with DNA. The vibe I got from the book was not one of inevitability, more "if Hammond had actually spared no expense on any single one of these dominos, the lot of them would not have fallen together." Admittedly it's been a while, but I specifically remember being miffed at the whole "chaos theory dictates that dinosaurs will find a way to escape every zoo" thing, which turned a morality story about cutting corners into a mash of nonsense. Chaos dictates no such thing, of course, but the nature of entertainment dictates that if the dinos don't get out then you and I won't pay to see the movie, so yeah, at the end of the day we really ought to be careful about what inferences we draw from entertainment where the conclusion is fixed. > Unfortunately another theme of Michael Crichton books is that because of human hubris and corporate greed, the disaster is inevitable. Regulation and scientific ethics don't matter because humans find a way. This is exactly what has happened with humanity. Another kind of cautionary tale could be that someone could, but doesn't advance something, because of moral reasons, and then someone else goes ahead and does it anyways. One of such tales is Ányos Jedlik, who might have had a lead in the creation of the electric motor and the dynamo, but we'll never truly know, as he was not the kind of person to go ahead and patent and build a business. Nevertheless, both were invented timely, and became the foundation of a more modern age. I think that at the end of the day, the dinos will get resurrected. What you can decide is what part you want of it. > What you can decide is what part you want of it. Or, you can build a revolution to take it down... I don't think that can happen. A revolution, and a new thing's concept are not in the same realm. Especially true if the thing has military implications. You just don't put that genie back to the bottle. Whatever do you, and I mean literally what ever, the thing will come to life. If it's socially acceptable, then the science magazines will hype about it, if not, it will be done in a laboratory that's not even on maps. We could take down the technological civilization, and institute methods more like the Amish...not saying it will happen like that but I mean that's certainly possible. I see no reason why all people would be on board, and so, I don't think that this is an angle that is worthy of exploration. I feel like it's the same utopian, or draconian kind of thinking that failed when implemented countless times in history. What you're right about is that we might try, nevertheless. I mean, we have things like the Geneva Convention or the United Nations, however debatable their effectiveness is. We wouldn't know unless we tried. But it's important to note that even though these things exist, violating them is always just an inch away. And then we're basically back to where I started: what you can decide is what part you want in the thing. But I don't think you can make the thing go away. But I mean, it's not that I invented this myself. It's basically the concept of Pandora's box. Have you any specific ideas? Would you mind sharing what sort of programming you have skills in? Or find a middle way. Call it the Butlerian Compromise. The later we postpone the greater will be disaster. Suppose we had this AI at the dawn of internet, we would have google using this tech to search internet back then. We would have saved countless hours of search to fix a bug. AI is having an Einstein in our pocket. Any one who had to develop/research on any topic, he has an Einstein level assistance ready to help you out. We are just at the dawn of the sci-fi era we always have imagined for. And AI will be our companion in this fast changing world: "Jarvis explain me how that thing works" , and you get somewhat of an idea about any new or complex subject, which otherwise would have taken days to comprehend. I think a "roadmap" should be also provided on how to learn this topic in a guided way. Also, some captions would be useful: what does it mean "model development"? Why "vector databases" ? What is "LLM Automation" ? What should I click first if I am a complete newbie ? I am not the Lazy programmer nor the owner of this website, but titular Lazy Programmer has developed several roadmaps of the courses they offer for those who wish to learn AI/ML. You don't necessarily have to take their courses, but you can glean the topics of the courses from their titles and get a good idea of what you need to know. The courses: https://lazyprogrammer.me/deep-learning-courses/ The order in which to take the courses: https://deeplearningcourses.com/course_order A video summary: Yes, this is a must! A newbies-focused roadmap would differentiate it from other linksites.
Newbies would benefit tremendously if they can grasp every keyword, and not get overwhelmed. This site could really be the starter platform of every AI enthusiast. I really hope that communities could build their own AI. For example, GPT-4-turbo doesn't get keyboard navigation for blind people right in just about any conversation. Why Microsoft made a chat bot for this very purpose of serving people with disabilities, I don't know. It's not even 100% accessible, the web app anyway. So, if we could train our own AI's, and multi-modal models, on how to help us, how to describe things, or even the kinds of media we each like, that'd be amazing! Is there something like this that is more broad than LLMs? There’s thousands of AI directories out there, from huggingface itself to the ones specialized in SaaSes Looks neat, kind of those "awesome X" list. Maybe adding a wiki would be interesting (so uninitiated can learn what is what and where to start if they want to build/train their own AI). Tangential question I am not sure where else to ask: Would it be possible to train AI at home on a book? I was thinking about how sometimes I start reading a book and then get distracted for a month or two. Summarizing chapters and characters up to a point in the book sounds like a good fit for GPT AI. I could possibly find and extract PDF or epub of the current book I am reading and I have RTX 4080 at home I am barely using. Is privateGPT what you're looking for? It stores PDFs in a vector DB and answers questions based on that content alone. Edit: Everything is local. This is not training, it takes a standard LLM and gives it extra context based on what text the tool finds in the vector DB related to the topic. That sounds even better than training LLM on my own. Thanks. Why wouldn't you just use Claude 3? It's going to be miles better than any OSS model. I'll have to read first what Claude 3 is. Thank you. Hey nice one, thanks for putting the links (along with GitHub repos) in a neat page for us! You build yahoo for AI. congratulations I guess How is this different than AI Awesome lists? Author here. Thanks for asking. One key objective is not to be yet another linksite, but add utility by providing direct shortcuts to code, docs, api references etc. next to the main link. > One key objective is not to be yet another linksite, but add utility by providing direct shortcuts to code, docs, api references etc. next to the main link. Is this a need people have? I would use letsbuild.ai to discover projects and check their homepage/documentation, but once I know a given project I know where to find its docs (or search for "<project> docs" on Google). don't underestimate the UX improvments from tiny friction removal features. It can become the difference between a site no one usess to a site everyone loves > don't underestimate the UX improvments from tiny friction removal features. It can become the difference between a site no one usess to a site everyone loves Which UX improvements are you thinking of exactly? This is a website I would personally not use because all I see is a list of links without context nor explanation about which tool is better than another for which task. This is great, thanks for making it! This is amazing, thank you! very informative,thanks!