No industry is going to be untouched by this. But like the introduction of the computer, there will be some industries that will be less impacted.
If I were a young man again or had sons that were teenagers, I would definitely be pursuing something in the trades such as construction or HVAC repair or nursing.
But I’m fairly certain many white collar jobs that can be done from a desk with a computer - e.g. engineering, coding, project management, graphic design - that are head skills and logic or creative based are going to be heavily consolidated or go away completely.
I’m just trying to hang on until retirement (not that retirement is guaranteed anymore).
The way GPT described itself reflects my own experience with these tools as well. It can take a while to "see" it, because the mimicry is good. But if you interact with them enough and in different contexts, you begin to see the "cracks" in the mimicry, as convincing as it can sometimes be, and you learn therefore to use it more appropriately.
I find GPT exceptionally useful at specific concrete tasks -- helping with image prompts (or any kind of AI prompts, for that matter -- something which is critical for these things to work well, and something which some people seem to find fascinating but which I find personally irritating, but which GPT can do for me in a heartbeat) and things like that. Getting into any in-depth discussion about anything "deep" is not what it's good at and very much "at your own risk?, in my opinion.
And the other limits all apply. Let's say you're talking to some chatbot who is playing the role of a companion -- like, say, a friend taking a stroll to a virtual park, it will easily lose track of geographic specifics (like where it is standing relative to that tree over there, or a lake or what have you) that are literally trivial for a human mind but absolutely flummoxing to an AI chatbot because, as GPT points out, it can't mimic that degree of specificity -- there is no probabilistic answer that will work almost all the time there, there's only the very specific one of "where is this specific tree" ... and it sucks at that. Because it isn't what it's doing ... and when you see the crack, you can't unsee it.
In fact, you start seeing that kind of crack everywhere you look once you know how to spot it. It's just that people are not used to working with these things so much (most people aren't anyway) and so they don't get to the point where they see the cracks, or they are too gullible about the often convincing early impression a chatbot can make, if it's well made.
--
I do think that the reasoning models are different. We've already seen the proto-versions of them, but they're very simple. In theory it *should* work, but in practice it takes mass amounts of compute to make the theory actually play out in a way that has been similar to what we have seen so far with LLMs ... and as we know, LLMs themselves take a truly massive amount of compute to work well. There are technical/hardware/capacity limitations that are coming into the picture, and I would not be terribly surprised if we reach some kind of a cap due to that sooner than we think. Not that I think this is coming soon, mind you, and I don't rule out the possibility that clever software or logic architecture design can stave off some of the capacity problems and limits, but I think we'll eventually get to them -- the elephant on the table, of course, is where the stuff will be when we get there (ie, how capable will it be).
I've only started using AI, at first I was hesitant.
it's a scary prospect as to what it will be like in the next 10-20 years especially in my line of work and whether I'll need to pivot to something else down the track.
Anyone with an in-depth knowledge of a particular topic can pretty quickly demonstrate the “garbage in, garbage out” nature of AI engines, particularly if there are opinions related to that topic that are dressed up as facts on popular websites. It’s a very efficient aggregator of information at this stage, but I wouldn’t want my life depending upon its accuracy.
No industry is going to be untouched by this. But like the introduction of the computer, there will be some industries that will be less impacted.
If I were a young man again or had sons that were teenagers, I would definitely be pursuing something in the trades such as construction or HVAC repair or nursing.
But I’m fairly certain many white collar jobs that can be done from a desk with a computer - e.g. engineering, coding, project management, graphic design - that are head skills and logic or creative based are going to be heavily consolidated or go away completely.
I’m just trying to hang on until retirement (not that retirement is guaranteed anymore).
The way GPT described itself reflects my own experience with these tools as well. It can take a while to "see" it, because the mimicry is good. But if you interact with them enough and in different contexts, you begin to see the "cracks" in the mimicry, as convincing as it can sometimes be, and you learn therefore to use it more appropriately.
I find GPT exceptionally useful at specific concrete tasks -- helping with image prompts (or any kind of AI prompts, for that matter -- something which is critical for these things to work well, and something which some people seem to find fascinating but which I find personally irritating, but which GPT can do for me in a heartbeat) and things like that. Getting into any in-depth discussion about anything "deep" is not what it's good at and very much "at your own risk?, in my opinion.
And the other limits all apply. Let's say you're talking to some chatbot who is playing the role of a companion -- like, say, a friend taking a stroll to a virtual park, it will easily lose track of geographic specifics (like where it is standing relative to that tree over there, or a lake or what have you) that are literally trivial for a human mind but absolutely flummoxing to an AI chatbot because, as GPT points out, it can't mimic that degree of specificity -- there is no probabilistic answer that will work almost all the time there, there's only the very specific one of "where is this specific tree" ... and it sucks at that. Because it isn't what it's doing ... and when you see the crack, you can't unsee it.
In fact, you start seeing that kind of crack everywhere you look once you know how to spot it. It's just that people are not used to working with these things so much (most people aren't anyway) and so they don't get to the point where they see the cracks, or they are too gullible about the often convincing early impression a chatbot can make, if it's well made.
--
I do think that the reasoning models are different. We've already seen the proto-versions of them, but they're very simple. In theory it *should* work, but in practice it takes mass amounts of compute to make the theory actually play out in a way that has been similar to what we have seen so far with LLMs ... and as we know, LLMs themselves take a truly massive amount of compute to work well. There are technical/hardware/capacity limitations that are coming into the picture, and I would not be terribly surprised if we reach some kind of a cap due to that sooner than we think. Not that I think this is coming soon, mind you, and I don't rule out the possibility that clever software or logic architecture design can stave off some of the capacity problems and limits, but I think we'll eventually get to them -- the elephant on the table, of course, is where the stuff will be when we get there (ie, how capable will it be).
I've only started using AI, at first I was hesitant.
it's a scary prospect as to what it will be like in the next 10-20 years especially in my line of work and whether I'll need to pivot to something else down the track.
Anyone with an in-depth knowledge of a particular topic can pretty quickly demonstrate the “garbage in, garbage out” nature of AI engines, particularly if there are opinions related to that topic that are dressed up as facts on popular websites. It’s a very efficient aggregator of information at this stage, but I wouldn’t want my life depending upon its accuracy.
It is, at minimum, more accurate than the majority of human beings, who all do essentially the same thing.