I'm not sure if it's intentional or not, but the Actually Useful AI community rarely gets any post on AI that's actually useful. Is this just a reflection of the popularity of this community or a reflection of the AI field in general?
Probably a little of both, but I really don't think there are very many practical applications for current AI technology. You can make bad drawings or bad writing in bulk, and those are both niches which have been overflowing with human contributors for centuries at least.
I also think somehow AI is getting less useful. The tighter they reign in what it's allowed to say and do and the more it sounds like a corporate tool, the less fun it is to try bending it to new purposes. We can draw and edit pictures, and we can summarize information, and we can write extremely simple computer code.
You can also make passable porn and smut, but the efforts to not let it do those things cripple it for other purposes such as tabletop roleplaying. I wrote a discord bot for this purpose almost a year also and I've spent way more time fighting with the prompt to allow combat or relationships than I have writing code and it's tedious. It's free to use in my private discord with a few select friends and no one has done anything with it in months. And these friends are all AI enthusiasts who became friends through AI Dungeon.
As an unregulated wild west frontier, AI is great and inspires creativity. As a sanitized corporate propaganda machine, it's really quite dull. I'm hoping some less censored versions catch up to OAI, but for now it feels like the gold rush is dying because it turned out to be nothing but pyrite.
That really depends on what you are calling AI. From my perspective, LLM, voice recognition and reproduction, and image generation, manipulation, and identification is the majority of AI. AGI doesn't exist outside of theory and primitive prototypes. If you want to include general programming using regex and case, that's not what I call AI.
Or are you saying the lion's share of AI is not on this list at all? Because I'd be interested to do more research on these other disciplines.
From my perspective, LLM, voice recognition and reproduction, and image generation, manipulation, and identification is the majority of AI.
I think that's a very short-sighted view of the subject. You're basically reducing AI to a very narrow set of specific, high-level practical applications of AI, leaving everything out of your personal definition.
The only thing I've deliberately left out is bog standard programming the likes of which I've done for 25 years. Which is why I asked what I'm missing.
Probably a little of both, but I really don’t think there are very many practical applications for current AI technology.
I don't think that's true at all.
A couple of days ago I read a post about someone using ChatGPT to generate commit messages from the diff. That's pretty cool.
Chatbots are also surprisingly helpful. Today's technology is already good enough to put together working expert systems. I'd love to have something like a chatbot that was fed something like all the papers from one or more academic journals related to a specialized field, and be able to ask it open-ended questions.
I think that nowadays the limit for actually useful AI is the imagination of those who work on today's cutting edge AI. It would be cool to hear more about those.
You are being silly and gullible. These chatbots just make shit up. You can't learn anything from them. It's literally worse than just googling the question and believing whatever nonsense you read on Quora. They don't "know" anything, they're a particularly advanced form of your phone's predictive text feature.
As someone who interacts with "these chatbots", the "shit" that they "make up" is very useful to onboard onto topics and find paths forward. They are the closest there is to expert systems, and currently there is no alternative that comes close to the value they provide.
You can’t learn anything from them.
Those who can't learn in general also won't learn in specifics.