Discussion
So keep reading about AI and how amazing it is. I do use ChatGTP to write things, like for our website or letters i find it can make things sound nice, all good.
I have tried to get it to make images but they are crap, am I doing something wrong, I presume it is related to the terms used and questions you ask of it or am I using the wrong platform?
For instance I asked it "can you put a 90s Lister storm car in front of the tower bridge" and it came up with :
I have tried to get it to make images but they are crap, am I doing something wrong, I presume it is related to the terms used and questions you ask of it or am I using the wrong platform?
For instance I asked it "can you put a 90s Lister storm car in front of the tower bridge" and it came up with :
nikaiyo2 said:
So keep reading about AI and how amazing it is. I do use ChatGTP to write things, like for our website or letters i find it can make things sound nice, all good.
I have tried to get it to make images but they are crap, am I doing something wrong, I presume it is related to the terms used and questions you ask of it or am I using the wrong platform?
For instance I asked it "can you put a 90s Lister storm car in front of the tower bridge" and it came up with :
Come on, that's not a bad attempt. I have tried to get it to make images but they are crap, am I doing something wrong, I presume it is related to the terms used and questions you ask of it or am I using the wrong platform?
For instance I asked it "can you put a 90s Lister storm car in front of the tower bridge" and it came up with :
Here's Gartners view
I think what we've seen over the last couple of years are tools that have been around for a while get investment, training on larger datasets and a public facing launch of tools like ChatGPT - hence the public perception that AI is progressing quickly.
My personal opinion, is that the general purpose tools we are seeing now, will likely get better at a progressively slower rate, with any obvious failings (i.e. 6 fingers!) getting targeted and tackled, but this will follow the law of diminishing returns - effort will be required to target each of these specific failings. The real growth in AI will be as targeted helpers for specific problems i.e. we are already seeing AI being rolled out in top-spec cameras to help identify subjects and focus correctly.
TLDR;
General purpose AI is not happening any time soon. Targetted AI for specific purposes will be endemic soon.
I think what we've seen over the last couple of years are tools that have been around for a while get investment, training on larger datasets and a public facing launch of tools like ChatGPT - hence the public perception that AI is progressing quickly.
My personal opinion, is that the general purpose tools we are seeing now, will likely get better at a progressively slower rate, with any obvious failings (i.e. 6 fingers!) getting targeted and tackled, but this will follow the law of diminishing returns - effort will be required to target each of these specific failings. The real growth in AI will be as targeted helpers for specific problems i.e. we are already seeing AI being rolled out in top-spec cameras to help identify subjects and focus correctly.
TLDR;
General purpose AI is not happening any time soon. Targetted AI for specific purposes will be endemic soon.
If you look into AI, you'll see that there's various different types, Machine Learning, Image Recognition, Neural Nets etc. etc.
Each of those works in a very specific way to do it's AI. And each can only do a specific set of things well. Image recognition for example can't write you a story.
Generative AI is the latest and greatest, but is only good for the one thing, generating words or pictures or music that is defined by it's training data.
We've managed to get Generative AI to a point where it'll output some pretty good stuff, decent pics and decent text. But that's all it'll do.
The problem is companies have invested trillions in the AI companies, and so they're now looking to get their return, so we're seeing Generative AI shoe-horned into as many areas as we can, whether it's the right fit or not.
General AI isn't generative AI, or machine learning or recognition. It's also not a "step" on from Generative AI, it's a whole new research area, and companies trying to do it are starting again at the beginning with their research. So while I'm not saying they can't come up with it, it's not a simple evolution of what we have and so unlikely to appear a few releases down the road.
Each of those works in a very specific way to do it's AI. And each can only do a specific set of things well. Image recognition for example can't write you a story.
Generative AI is the latest and greatest, but is only good for the one thing, generating words or pictures or music that is defined by it's training data.
We've managed to get Generative AI to a point where it'll output some pretty good stuff, decent pics and decent text. But that's all it'll do.
The problem is companies have invested trillions in the AI companies, and so they're now looking to get their return, so we're seeing Generative AI shoe-horned into as many areas as we can, whether it's the right fit or not.
General AI isn't generative AI, or machine learning or recognition. It's also not a "step" on from Generative AI, it's a whole new research area, and companies trying to do it are starting again at the beginning with their research. So while I'm not saying they can't come up with it, it's not a simple evolution of what we have and so unlikely to appear a few releases down the road.
It's lots of fun having a poke at the current generation of AI hype driven stuff, everyone is so keen to show they're using the latest shiny even if it's less appropriate for the job than an older AI technique or even no AI at all.
The easy part is writing a paper or doing a presentation or running a demo, a lot of it breaks down very rapidly if poked or will just never be capable of much. Always makes a Q & A good fun especially if you're feeling vindictive (having been wound up with a shiny 'it works!' demo of something they just deeply failed to deliver in working form due to fundamental conceptual flaws).
The biggest ugly secret is the sheer amount of manual cludgery that's lurking behind the curtain to keep all that clever automated machine learning on the right path; a lot of it only sort of works, and plausible != real.
There's so much bulls
t and the sad bit is a lot of what people want to do *is* achievable, just not the way they want to do it. They're more eager to stay aligned to the hype than to get something working.
Also maybe doesn't help (in my experience anyway) that too much is maths & physics - and marketing - types pottering about in Python rather than engineer types creating something to do a job; they're happy that it does *something*, and not too worried that the 'something' either isn't right or it's a worse way to do it than other existing methods.
Best fun of all - we can train you a model on data. What data? How are you going to clean it or classify it or get enough coverage to be more useful than the existing models? And you want how long & how much?!
TL;DR - too much hype & bullst in AI land, bane of my life trying to keep control of AI related projects & products, AI specialists need firm guidance (and sometimes calling out on their misunderstanding of their own subject) if you want something useful from it instead of just something that's good for conferences/shows/demos and that falls apart if you wanted to use it in anger or sell it. If you want the useful AI it's probably not going to be from a massive pattern generator or a chatbot that distracts from all the actual good stuff.
TL;DR² - AI; the latest GPT stuff is mostly useless hype driven bullst chasing money & purpose, and unlikely to move from that to truly worthwhile except for the cases where a generic picture or chat or plausible-ish content is all you want.
TL;DR³ - fking AI...
The easy part is writing a paper or doing a presentation or running a demo, a lot of it breaks down very rapidly if poked or will just never be capable of much. Always makes a Q & A good fun especially if you're feeling vindictive (having been wound up with a shiny 'it works!' demo of something they just deeply failed to deliver in working form due to fundamental conceptual flaws).
The biggest ugly secret is the sheer amount of manual cludgery that's lurking behind the curtain to keep all that clever automated machine learning on the right path; a lot of it only sort of works, and plausible != real.
There's so much bulls
t and the sad bit is a lot of what people want to do *is* achievable, just not the way they want to do it. They're more eager to stay aligned to the hype than to get something working. Also maybe doesn't help (in my experience anyway) that too much is maths & physics - and marketing - types pottering about in Python rather than engineer types creating something to do a job; they're happy that it does *something*, and not too worried that the 'something' either isn't right or it's a worse way to do it than other existing methods.
Best fun of all - we can train you a model on data. What data? How are you going to clean it or classify it or get enough coverage to be more useful than the existing models? And you want how long & how much?!
TL;DR - too much hype & bulls
t in AI land, bane of my life trying to keep control of AI related projects & products, AI specialists need firm guidance (and sometimes calling out on their misunderstanding of their own subject) if you want something useful from it instead of just something that's good for conferences/shows/demos and that falls apart if you wanted to use it in anger or sell it. If you want the useful AI it's probably not going to be from a massive pattern generator or a chatbot that distracts from all the actual good stuff.TL;DR² - AI; the latest GPT stuff is mostly useless hype driven bulls
t chasing money & purpose, and unlikely to move from that to truly worthwhile except for the cases where a generic picture or chat or plausible-ish content is all you want. TL;DR³ - f
king AI...With luck the inevitable implosion of the hype into a small puff of 'meh' is just around the corner - they can deny gravity for a while but it wins in the end.
Ultimately however much it's hyped and pushed, and however much has been invested so far, if it's becoming exponentially more expensive to stay in the game with datacentres & models, and stuff still only mostly works, and prior investment in hardware & code depreciates to nothing too quickly and (the big one!) there isn't the demand or paying customers to keep it afloat at the scale some expect, then it's going to wither & die.
The thing is a nice toy and it has some genuine uses but you can smell the desperation from certain quarters to get people to please please use it, and (maybe) one day pay for it.
Tech hype comes, tech hype goes, billions is incinerated, the world moves on. The useful bits will carry on as indeed they did *before* the hype started. AI will go back to being just another tool with the current high profile LLM bits becoming another part of the background like all the other clever AI stuff no-one notices every day.
It's not like we haven't had the same cycle so many times before...
Ultimately however much it's hyped and pushed, and however much has been invested so far, if it's becoming exponentially more expensive to stay in the game with datacentres & models, and stuff still only mostly works, and prior investment in hardware & code depreciates to nothing too quickly and (the big one!) there isn't the demand or paying customers to keep it afloat at the scale some expect, then it's going to wither & die.
The thing is a nice toy and it has some genuine uses but you can smell the desperation from certain quarters to get people to please please use it, and (maybe) one day pay for it.
Tech hype comes, tech hype goes, billions is incinerated, the world moves on. The useful bits will carry on as indeed they did *before* the hype started. AI will go back to being just another tool with the current high profile LLM bits becoming another part of the background like all the other clever AI stuff no-one notices every day.
It's not like we haven't had the same cycle so many times before...
Gassing Station | Computers, Gadgets & Stuff | Top of Page | What's New | My Stuff