Regarding AI and machine learning, have you come across the software development that was taught to pick up cancer tumours? It’s standard practice for doctors to put a ruler alongside a tumour to show how big it is. The software’s ability to detect tumours ended up being the sight of a ruler in the photo 🤣🤣 so rulers cause cancer
My main problem with the question of AI as you’ve detailed it is that your point is 100% correct, very well documented and unlikely to change anytime soon, and yet none of this is stopping these absolute fucktangles from presenting said glorified spreadsheets as alternatives to search engines or, you know, actual people with expertise. There are teachers tearing out what little hair they have because a whole generation of kids doesn’t know the difference between an AI chatbot and a search engine (which, in fairness, are also not particularly reliable these days, but that’s another head of the hydra of late-stage capitalism).
My day job is writing clinical and technical content for the internal and external knowledge bases of a large UK insurer. At present we are training two AI chatbots (one for internal, agent-facing content, one for external customer-facing content) to deliver correct responses on complex queries. We have already determined that the core issues these bots are facing in delivering said correct responses could not have been predicted by us prior to testing. In other words, ONLY testing can flag these core issues so that we can fix them, either by adjusting the content so that the bot is happy, or retraining the bot.
However, since a knowledge base is an organic, fluid construction which is constantly being updated and built upon, there will always be content that the AI needs training on. Since we can’t be sure that new or updated content will not spark a juddering fit and a wave of incorrect answers without testing, it follows that all new and updated content will need running past the chatbot and testing before we publish it.
I’m not sure how to get people in authority to understand this, because at present everyone I’m speaking to seems to be under the impression that these AIs are supposed to *know* the answers after having access to properly calibrated content. Explaining that AI chatbots are only supposed to *sound like* they know the answers is meeting with uncomfortable glances and/or baffled stares…
They used to teach that in any project the three metrics you needed to worry about were TIME, COST and QUALITY. So when some well-meaning idiot wanted to move the project scope, you had a discussion around whether this new requirement should be accommodated through flexing (a) TIME (give us an extra month to deliver), (b) COST (give us more money so we can get more people and stuff on the case) or (c) QUALITY (are you willing to accept a shittier end outcome?) The answer was always (c) because (c) gets the box ticked. AI gives unlimited scope to reduce TIME and COST and there are enough people who don't give a fuck about QUALITY to make it irresistible.
My daughter's husband works on developing AI and tells a great story. He asked it to write him a paper on a subject that he happens to be an expert on, and it quickly generated a pretty good paper. He asked it to cite its sources and it happily regurgitated. Only thing is- the sources it cited do not exist. It called for papers by people he personally knew that had not been written.
AI is far more limited than all these starry eyed techbros claim, and that is dangerous considering how eager people are to hand the controls of everything over to HAL-9000.
As for the Lost Causes... well, sadly the left end of the bell curve will always exist. Or perhaps we should call it the belm curve.
I think we should be worried about AI, but purely because of the human factor in it in one way or another. Developing it, worshipping it or relying on it, we are bound to screw it up one way or another. Also - there are three “Straw Men” novels?!? Gasp! How did I miss that? I bet if I asked AI it would have told me. I got pretty traumatized by the end of the first one. Maybe I didn’t want to know? Will have to seek it out now and see how the things turned out. :)
Of course they do :) I blame Twitter. I was minding my own business (not really) where I came across Mr. Smith with his eloquent verse and relatable jokes and even more relatable grumpiness… so finally I just had to start reading the books, which did not disappoint, but they sure freak me out on a regular basis… which is what I need, it seems, to achieve equilibrium. So it all works out in the end. My favorite so far is “When God Lived In Kentish Town”. But, yeah… “Straw Men” has nowhere to go but to the “read under blanket with all the lights on and triple check the locks before you start” territory.
How does one “read them wrong”?
I seem to read “Only forward” wrong. Don’t seem to be able to get past the first two pages and now I misplaced the book… that might be the wrong way to read :) Maybe I’ll start reading from second chapter next time.
Oh, it’s fine. I am an omnivore. Different is good… Need to find it first.
Do you have a favorite?
Read The Intruders. I think I found Straw Men resolution more surprising, also more heartbreaking and scary and more messed up. Messed up as in “someone needs to pick up all those pieces, label them, catalogue them and put them back on the shelf” Did get the second book last night. We’ll see.
I also have We Are Here and Time Out waiting their turn.
Books must be very patient creatures indeed. They just sit there thinking: “It’s alright, darling, take your time. I’ll get you in the end”. They get authors first and then just keep moving from mind to mind forever… growing, evolving, inspiring new books and starting the process all over again”. Books are really just ancient monsters looking for a mind to consume.
I recently saw a couple of posts from Paul Schrader, of Taxi Driver fame, praising story ideas he requested from ChatGPT. He went so far as to say that AI has "better ideas" than him and questioning why a writer should sit around and think for months, when a bot can come up with something "in seconds".
I was a little shocked at this, given his career, but perhaps he is a tad burnt out. The idea that he would want to skip the joy of creation, of coming up with something of his own, in favour of being handed something he just has to tweak, suggests maybe he's just a bit tired of the job and is simply going through the motions.
I really hope this attitude doesn't become prevalent among established writers.
Yes, that sounds like burn-out to me... and/or a guy who's so weary of dredging around in his own head, and feels like he's seen everything he can come up with, and is desperate for some "other" in his inspiration.
One would think he might simply retire at that point, but I get the impression that retirment in Hollywood, where you're often only as good as your last hit, is somewhat different to retirement in the rest of the world.
I’m being hit with ads from an app called Blinkist whose USP seems to be that it condenses books into 5min reads! I’m assuming it uses AI to do the butchery. And I’m guessing from what you’ve explained, how it’s terrible at doing that, that many people will not get the meaning of a book as it is intended by the author, not that they would in 5mins anyway. It seems barbaric and just.. stupid. Why would you do that to a book! It’s meant to be read. I don’t like it.
As someone who has worked with machine translations (ie semi-AI) for a long time I don’t think AI is anywhere near being reliable to do any work without supervision - they do however make us work faster if we use them correctly (and don’t cut corners so we have to go back and correct stupid mistakes it makes).
We were in Scotland during the Merrineum (TM and C whoever said that first on The Other Place). at the start of the recent cold snap, in our campervan on the shores of Loch Ness, in the grounds of Scone Palace in Perth, and, yes, in Edinburgh with tickets to the cancelled Hogmanay street party...
The drive down The Great Glen to Fort William, up and over Glencoe and Rannoch Moor and back via Lomond and the Trossachs, in the first few inches of snow and hard frost was really quite a special way to start the new year.
Yes, I know I'm missing the point(s) of your post, but here we are.
Yes, I think that's true... and I definitely find her (I think of it the same way) useful on day to day quick questions and tasks... more like a first draft research assistant...
I'm worried that several generations have not been taught critical thinking skills and aren't able to determine that AI is just flat-out wrong a lot of the time (but infinitely confident despite being wrong).
I'm also worried about the enormous amount of power and water these huge AI systems are using and the fact that tech companies are starting to admit that AI can never be profitable unless they're allowed to steal all content from the people who create it.
There are tricks to using gen AI that a team at eDiscovery company Merlin Technology worked out. They have a limited context window, i.e. they can only handle a few thousand words at once. If you can get it o start with an outline, then have it expand each each section (while keeping a close eye out for hallucinations). They’ve had success with that approach.
Regarding AI and machine learning, have you come across the software development that was taught to pick up cancer tumours? It’s standard practice for doctors to put a ruler alongside a tumour to show how big it is. The software’s ability to detect tumours ended up being the sight of a ruler in the photo 🤣🤣 so rulers cause cancer
Brilliant.
My main problem with the question of AI as you’ve detailed it is that your point is 100% correct, very well documented and unlikely to change anytime soon, and yet none of this is stopping these absolute fucktangles from presenting said glorified spreadsheets as alternatives to search engines or, you know, actual people with expertise. There are teachers tearing out what little hair they have because a whole generation of kids doesn’t know the difference between an AI chatbot and a search engine (which, in fairness, are also not particularly reliable these days, but that’s another head of the hydra of late-stage capitalism).
My day job is writing clinical and technical content for the internal and external knowledge bases of a large UK insurer. At present we are training two AI chatbots (one for internal, agent-facing content, one for external customer-facing content) to deliver correct responses on complex queries. We have already determined that the core issues these bots are facing in delivering said correct responses could not have been predicted by us prior to testing. In other words, ONLY testing can flag these core issues so that we can fix them, either by adjusting the content so that the bot is happy, or retraining the bot.
However, since a knowledge base is an organic, fluid construction which is constantly being updated and built upon, there will always be content that the AI needs training on. Since we can’t be sure that new or updated content will not spark a juddering fit and a wave of incorrect answers without testing, it follows that all new and updated content will need running past the chatbot and testing before we publish it.
I’m not sure how to get people in authority to understand this, because at present everyone I’m speaking to seems to be under the impression that these AIs are supposed to *know* the answers after having access to properly calibrated content. Explaining that AI chatbots are only supposed to *sound like* they know the answers is meeting with uncomfortable glances and/or baffled stares…
Oy vey. There are hard times ahead, for humans whose job it apparently now is to prove that non-humans to do jobs that... they can't.
They used to teach that in any project the three metrics you needed to worry about were TIME, COST and QUALITY. So when some well-meaning idiot wanted to move the project scope, you had a discussion around whether this new requirement should be accommodated through flexing (a) TIME (give us an extra month to deliver), (b) COST (give us more money so we can get more people and stuff on the case) or (c) QUALITY (are you willing to accept a shittier end outcome?) The answer was always (c) because (c) gets the box ticked. AI gives unlimited scope to reduce TIME and COST and there are enough people who don't give a fuck about QUALITY to make it irresistible.
VERY true. And reminds me of a saying from my graphic design days: "This job can be cheap, quick, or good. But you can only have two of them."
My daughter's husband works on developing AI and tells a great story. He asked it to write him a paper on a subject that he happens to be an expert on, and it quickly generated a pretty good paper. He asked it to cite its sources and it happily regurgitated. Only thing is- the sources it cited do not exist. It called for papers by people he personally knew that had not been written.
AI is far more limited than all these starry eyed techbros claim, and that is dangerous considering how eager people are to hand the controls of everything over to HAL-9000.
As for the Lost Causes... well, sadly the left end of the bell curve will always exist. Or perhaps we should call it the belm curve.
Or as we say in the UK: The Bell-end Curve ;-)
I probably shouldn't admit to knowing this, but...
https://www.urbandictionary.com/define.php?term=Belm
I think we should be worried about AI, but purely because of the human factor in it in one way or another. Developing it, worshipping it or relying on it, we are bound to screw it up one way or another. Also - there are three “Straw Men” novels?!? Gasp! How did I miss that? I bet if I asked AI it would have told me. I got pretty traumatized by the end of the first one. Maybe I didn’t want to know? Will have to seek it out now and see how the things turned out. :)
The Straw Men series, like this substack, only gets more disturbing and visceral.
Although I may be reading them wrong...
Of course they do :) I blame Twitter. I was minding my own business (not really) where I came across Mr. Smith with his eloquent verse and relatable jokes and even more relatable grumpiness… so finally I just had to start reading the books, which did not disappoint, but they sure freak me out on a regular basis… which is what I need, it seems, to achieve equilibrium. So it all works out in the end. My favorite so far is “When God Lived In Kentish Town”. But, yeah… “Straw Men” has nowhere to go but to the “read under blanket with all the lights on and triple check the locks before you start” territory.
How does one “read them wrong”?
I seem to read “Only forward” wrong. Don’t seem to be able to get past the first two pages and now I misplaced the book… that might be the wrong way to read :) Maybe I’ll start reading from second chapter next time.
OF is quite a different book to the SM ones... you might be better off with something like THE INTRUDERS...
Oh, it’s fine. I am an omnivore. Different is good… Need to find it first.
Do you have a favorite?
Read The Intruders. I think I found Straw Men resolution more surprising, also more heartbreaking and scary and more messed up. Messed up as in “someone needs to pick up all those pieces, label them, catalogue them and put them back on the shelf” Did get the second book last night. We’ll see.
I also have We Are Here and Time Out waiting their turn.
Books must be very patient creatures indeed. They just sit there thinking: “It’s alright, darling, take your time. I’ll get you in the end”. They get authors first and then just keep moving from mind to mind forever… growing, evolving, inspiring new books and starting the process all over again”. Books are really just ancient monsters looking for a mind to consume.
Oh yes, there's more! If you try them out, let me know what you think :-)
I recently saw a couple of posts from Paul Schrader, of Taxi Driver fame, praising story ideas he requested from ChatGPT. He went so far as to say that AI has "better ideas" than him and questioning why a writer should sit around and think for months, when a bot can come up with something "in seconds".
I was a little shocked at this, given his career, but perhaps he is a tad burnt out. The idea that he would want to skip the joy of creation, of coming up with something of his own, in favour of being handed something he just has to tweak, suggests maybe he's just a bit tired of the job and is simply going through the motions.
I really hope this attitude doesn't become prevalent among established writers.
Yes, that sounds like burn-out to me... and/or a guy who's so weary of dredging around in his own head, and feels like he's seen everything he can come up with, and is desperate for some "other" in his inspiration.
One would think he might simply retire at that point, but I get the impression that retirment in Hollywood, where you're often only as good as your last hit, is somewhat different to retirement in the rest of the world.
I’m being hit with ads from an app called Blinkist whose USP seems to be that it condenses books into 5min reads! I’m assuming it uses AI to do the butchery. And I’m guessing from what you’ve explained, how it’s terrible at doing that, that many people will not get the meaning of a book as it is intended by the author, not that they would in 5mins anyway. It seems barbaric and just.. stupid. Why would you do that to a book! It’s meant to be read. I don’t like it.
Me neither. it makes a fool of both reader and writer. Bah.
As someone who has worked with machine translations (ie semi-AI) for a long time I don’t think AI is anywhere near being reliable to do any work without supervision - they do however make us work faster if we use them correctly (and don’t cut corners so we have to go back and correct stupid mistakes it makes).
Yes, you're right, like any tool - it'll be a case of learning how to use them...
Welcome home(ish)!
We were in Scotland during the Merrineum (TM and C whoever said that first on The Other Place). at the start of the recent cold snap, in our campervan on the shores of Loch Ness, in the grounds of Scone Palace in Perth, and, yes, in Edinburgh with tickets to the cancelled Hogmanay street party...
The drive down The Great Glen to Fort William, up and over Glencoe and Rannoch Moor and back via Lomond and the Trossachs, in the first few inches of snow and hard frost was really quite a special way to start the new year.
Yes, I know I'm missing the point(s) of your post, but here we are.
That sounds fabulous. I've not spent anywhere near enough time in Scotland...
Scotland. Is it cold? Then you use it! IA… Let's see.….
It's chilly, but not FREEZING ;-)
You can ask ChatGPT about adjusting its settings to reduce hallucinations. :)
ChatGPT has saved me countless hours of tedious work. I think of her (I identify ChatGPT as female) as a secretary with a lot of personality.
As with any tool, make sure not to use ChatGPT for tasks it isn’t designed to handle.
Yes, I think that's true... and I definitely find her (I think of it the same way) useful on day to day quick questions and tasks... more like a first draft research assistant...
I'm worried that several generations have not been taught critical thinking skills and aren't able to determine that AI is just flat-out wrong a lot of the time (but infinitely confident despite being wrong).
I'm also worried about the enormous amount of power and water these huge AI systems are using and the fact that tech companies are starting to admit that AI can never be profitable unless they're allowed to steal all content from the people who create it.
Yep... and nobody's going to do anything about it.
There are tricks to using gen AI that a team at eDiscovery company Merlin Technology worked out. They have a limited context window, i.e. they can only handle a few thousand words at once. If you can get it o start with an outline, then have it expand each each section (while keeping a close eye out for hallucinations). They’ve had success with that approach.
Ohhhhh interesting. Thank you!