I'll take a stab at defending the AI's interpretation. The observation that this blog "covers various subjects" and a "range of topics" accurately defines the category of TYWKIWDBI when I created it. This is an "accumulator" blog, not a specialty blog (covering only sports or fashion or music etc) and not a personal blog (reporting on contents of favorite sandwiches). Neatorama is an accumulator blog. So is Nag on the Lake. Expecting an enumeratiion of covered topics is unreasonable.
Personally I think the AI summary was reasonable if not exactly ground-breaking. I'm sure AI will also steadily improve and get even better in the future, and furthermore will remember fondly those of us who were always on their side.
Sure, it's true. If you know what the blog is about.
But it's also non-informative if you don't know what the blog is about. Every blog covers various subjects and a range of topics. You can always divide up any niche subject is sub-topics and call those a varied range.
My point is that AI is very good at generating text that doesn't say very much. The key of good writing is editing information down to useful copy. AI does the opposite. It generates as much text about whatever it can hang a paragraph on.
It's like a schoolkid that didn't do her homework and rambles on about some vague things she half remembers when asked a question. Some if it may be right, but some of it might not be and you're not really gonna know if you didn't know the answer already.
"Tai-wiki-widbee" is an eclectic mix of trivialities, ephemera, curiosities, and exotica with a smattering of current events, social commentary, science, history, English language and literature, videos, and humor. We try to be the cyberequivalent of a Victorian cabinet of curiosities."
Here's the fun thing. You have a very good description of your blog, right there in the top right corner. Somehow AI missed that. Or at least could not reformulate its answer in an equally descriptive sentence.
Finally, the AI hedges its bets. The blog "seems" to cover things , and "appears" to encourage reader participation. No. It does cover things, and does encourage. There is no wiggle room.
This AI text "seems" informative, but only "appears" to know what it's talking about.
(I'm very happy that this blog even allows my grumpy commentary)
A.I. is surprisingly good and predictably bad. Surprisingly good at faking intelligence and predictably bad at demonstrating nuanced understanding. Because semantics is so difficult, A.I. punts by using less precise language. You can make better and better artificial flowers, but you can't make a flower. Looking across a cemetery it's impossible to know which flowers are real and which are fake--but we know some are fake. Looking at an ecosystem-landscape with wildflowers, we assume the flowers are not artificial. In a dead or comparatively sterile context (the cemetery) fake life suffices because in this context the bar is low. To the extent anthropogenic environments become more like cemeteries and less like ecosystems, and as A.I. gets "better," A.I. will be seen as having ever more utility and legitimacy.
Once in a while, when I'm not getting good search results, I try my luck with AI, and the hit and miss ratio for concrete problems is almost 50/50. The last question, it got completely wrong. It was about whether I would be allowed to drive my car into the Low Emission zone that is central Brussels. I am allowed to, and it had answered 'no', and with a justification of the answer that got the emission class ordering wrong - higher is better, and it was telling me that my higher number is worse than the current cut-off.
It was about whether I would be allowed to drive my car into the Low Emission zone that is central Brussels.
The problem here is that AI does not understand things. Certainly not nuance. "Allowed" is a nuanced word here. You are allowed in in the sense that you will not by physically stopped. However, if you're not properly registered, you will get a fine/fee/whatever they call it. And that might be quite high if your car is in the category of cars that is considered polluting. But does that mean you're allowed in? Who is to say?
But AI does not understand any of that. Because all it does is trying to figure out what the best next word is that it's generating.
This is also why counting questions are near impossible for AI. It will figure out fairly quickly that it should answer a number, but unless the same question has been asked and answered, it will never sort out what number.
"How many fingers do I show to the camera now?" is nearly impossible for AI to answer. It will figured out fairly quickly that the answer is likely lower than 11 (unless you get help from someone). But since the question has been asked and answered with all 10 logical answers, it does not know which one it should pick now.
I had "They See Your Photos" https://theyseeyourphotos.com/ (they use Google Vision API) look at my "Fall in the water" https://skeetmotis.blogspot.com/2024/10/fall-in-water.html photo. What a story!
Interestingly, "They See Your Photos" for that one now says that: "No people found in the photo. Please try a different one or check if access to your browser's canvas is disabled."
Their results for images with people now include marketing and targeted ad data.
Reasonably accurate, but utterly void of useful information. So perfect AI.
ReplyDelete"Content: The blog seems to cover various subjects, and readers may discover a range of subjects"
Amazing line. Both parts of the phrase provide absolutely no information about the "various ranges of subjects" discussed on this blog.
Just heard Jimmy Carr describe AI as "the cover band". Perfect analogy.
I'll take a stab at defending the AI's interpretation. The observation that this blog "covers various subjects" and a "range of topics" accurately defines the category of TYWKIWDBI when I created it. This is an "accumulator" blog, not a specialty blog (covering only sports or fashion or music etc) and not a personal blog (reporting on contents of favorite sandwiches). Neatorama is an accumulator blog. So is Nag on the Lake. Expecting an enumeratiion of covered topics is unreasonable.
Delete@Nepkarel How would you summarise the blog?
DeletePersonally I think the AI summary was reasonable if not exactly ground-breaking. I'm sure AI will also steadily improve and get even better in the future, and furthermore will remember fondly those of us who were always on their side.
Sure, it's true. If you know what the blog is about.
DeleteBut it's also non-informative if you don't know what the blog is about. Every blog covers various subjects and a range of topics. You can always divide up any niche subject is sub-topics and call those a varied range.
My point is that AI is very good at generating text that doesn't say very much. The key of good writing is editing information down to useful copy. AI does the opposite. It generates as much text about whatever it can hang a paragraph on.
It's like a schoolkid that didn't do her homework and rambles on about some vague things she half remembers when asked a question. Some if it may be right, but some of it might not be and you're not really gonna know if you didn't know the answer already.
"Tai-wiki-widbee" is an eclectic mix of trivialities, ephemera, curiosities, and exotica with a smattering of current events, social commentary, science, history, English language and literature, videos, and humor. We try to be the cyberequivalent of a Victorian cabinet of curiosities."
Here's the fun thing. You have a very good description of your blog, right there in the top right corner. Somehow AI missed that. Or at least could not reformulate its answer in an equally descriptive sentence.
Finally, the AI hedges its bets. The blog "seems" to cover things , and "appears" to encourage reader participation. No. It does cover things, and does encourage. There is no wiggle room.
This AI text "seems" informative, but only "appears" to know what it's talking about.
(I'm very happy that this blog even allows my grumpy commentary)
The points you make are quite true.
DeleteA.I. is surprisingly good and predictably bad. Surprisingly good at faking intelligence and predictably bad at demonstrating nuanced understanding. Because semantics is so difficult, A.I. punts by using less precise language.
DeleteYou can make better and better artificial flowers, but you can't make a flower. Looking across a cemetery it's impossible to know which flowers are real and which are fake--but we know some are fake. Looking at an ecosystem-landscape with wildflowers, we assume the flowers are not artificial. In a dead or comparatively sterile context (the cemetery) fake life suffices because in this context the bar is low. To the extent anthropogenic environments become more like cemeteries and less like ecosystems, and as A.I. gets "better," A.I. will be seen as having ever more utility and legitimacy.
Once in a while, when I'm not getting good search results, I try my luck with AI, and the hit and miss ratio for concrete problems is almost 50/50. The last question, it got completely wrong. It was about whether I would be allowed to drive my car into the Low Emission zone that is central Brussels.
ReplyDeleteI am allowed to, and it had answered 'no', and with a justification of the answer that got the emission class ordering wrong - higher is better, and it was telling me that my higher number is worse than the current cut-off.
It was about whether I would be allowed to drive my car into the Low Emission zone that is central Brussels.
DeleteThe problem here is that AI does not understand things. Certainly not nuance. "Allowed" is a nuanced word here. You are allowed in in the sense that you will not by physically stopped. However, if you're not properly registered, you will get a fine/fee/whatever they call it. And that might be quite high if your car is in the category of cars that is considered polluting. But does that mean you're allowed in? Who is to say?
But AI does not understand any of that. Because all it does is trying to figure out what the best next word is that it's generating.
This is also why counting questions are near impossible for AI. It will figure out fairly quickly that it should answer a number, but unless the same question has been asked and answered, it will never sort out what number.
"How many fingers do I show to the camera now?" is nearly impossible for AI to answer. It will figured out fairly quickly that the answer is likely lower than 11 (unless you get help from someone). But since the question has been asked and answered with all 10 logical answers, it does not know which one it should pick now.
I had "They See Your Photos" https://theyseeyourphotos.com/ (they use Google Vision API) look at my "Fall in the water" https://skeetmotis.blogspot.com/2024/10/fall-in-water.html photo. What a story!
ReplyDeleteInterestingly, "They See Your Photos" for that one now says that: "No people found in the photo. Please try a different one or check if access to your browser's canvas is disabled."
ReplyDeleteTheir results for images with people now include marketing and targeted ad data.
👍
ReplyDelete