3 ms·

Coming from someone where AI has not been one of my top interests for the last 30 years, what specifically should the focus be? In the last several months I ha

Coming from someone where AI has not been one of my top interests for the last 30 years, what specifically should the focus be? In the last several months I have used ChatGPT for things like thought experiments and churning out code faster.

mindcrime 3y ago
Coming from someone where AI has not been one of my top interests for the last 30 years, what specifically should the focus be? That's tough to answer. I have some specific areas of interest that I've cultivated over years and decades of maintaining a somewhat passive interest in AI, and only starting to really work on more serious research over the past few years. And even then, my list of "stuff to look into" is somewhat speculative, so I'm not sure how much value there would be in going into it in detail. That said, I've alluded to this briefly in previous HN posts, and it's not any kind of secret or anything. My interest areas include a heavy-dose of focus on "neuro-symbolic" systems: that is, integrating neural networks and "symbolic computing" (aka "GOFAI" or "Good Old Fashioned AI). To drill down a bit more: the idea is to use neural networks where they are really good at pattern recognition: image recognition, speech recognition, some language understanding tasks, etc., and then integrate that with more purpose built systems that work at a "symbolic level" (or you might say "knowledge level") to do things like deductive reasoning, abductive inference, temporal reasoning, inductive inference, common sense reasoning, etc. One of the tripping points for this is how to translate between the representations of "knowledge" across those different modalities. In the last several months I have used ChatGPT for things like thought experiments and churning out code faster. Yeah, I think for anybody who maybe isn't interested in diving into AI research per-se, doing stuff like that is very valuable. At the very least, learning to use and manage and work with the AI tools/platforms that are put out there will probably be an increasingly valuable skill over the years to come. The only thing I'd add to this is to say that I think it might be a mistake to focus only on LLM's. Don't get me wrong: I'm amazed at what LLM's can do and I think they are - and will continue to be - very important, but I don't know that I think they are the "be all, end all" of AI. It might be worthwhile to look at the suite of AI/ML offerings being provided by cloud providers like AWS, Azure, GCP, etc. and at least get familiar with working with some of those things.

mindcrime 3y ago

Coming from someone where AI has not been one of my top interests for the last 30 years, what specifically should the focus be? That's tough to answer. I have some specific areas of interest that I've cultivated over years and decades of maintaining a somewhat passive interest in AI, and only starting to really work on more serious research over the past few years. And even then, my list of "stuff to look into" is somewhat speculative, so I'm not sure how much value there would be in going into it in detail. That said, I've alluded to this briefly in previous HN posts, and it's not any kind of secret or anything. My interest areas include a heavy-dose of focus on "neuro-symbolic" systems: that is, integrating neural networks and "symbolic computing" (aka "GOFAI" or "Good Old Fashioned AI). To drill down a bit more: the idea is to use neural networks where they are really good at pattern recognition: image recognition, speech recognition, some language understanding tasks, etc., and then integrate that with more purpose built systems that work at a "symbolic level" (or you might say "knowledge level") to do things like deductive reasoning, abductive inference, temporal reasoning, inductive inference, common sense reasoning, etc. One of the tripping points for this is how to translate between the representations of "knowledge" across those different modalities. In the last several months I have used ChatGPT for things like thought experiments and churning out code faster. Yeah, I think for anybody who maybe isn't interested in diving into AI research per-se, doing stuff like that is very valuable. At the very least, learning to use and manage and work with the AI tools/platforms that are put out there will probably be an increasingly valuable skill over the years to come. The only thing I'd add to this is to say that I think it might be a mistake to focus only on LLM's. Don't get me wrong: I'm amazed at what LLM's can do and I think they are - and will continue to be - very important, but I don't know that I think they are the "be all, end all" of AI. It might be worthwhile to look at the suite of AI/ML offerings being provided by cloud providers like AWS, Azure, GCP, etc. and at least get familiar with working with some of those things.