I've written a few posts here and there over the years. I've never used AI and have no plans to do so. I kinda like picking my own words, doing my own research, thinking my own thoughts and expressing them in whatever way I choose. But then, I'm old. On the other hand, whatever I write is mine, for better or worse.
One of my favorite use cases is generating recipes for dinner. I list a bunch of ingredients I want to use up or are about to go bad and it almost always comes up with something decent. It's especially good at making very specific substitutions: "rewrite that chicken stir fry recipe to use canned chicken instead of fresh chicken breasts" for example.
Yes! I also used it to get feedback on how to optimize familiar recipes if I'm interested in streamlining the process (e.g. when should I drop in greens? how should I avoid splatter? etc).
That Python thing sounds like the kind of thing you should try and monetize. A very niche problem with a solution that saves a ton of time while also vastly improving usability. Deserves to make money!
Recently, after watching /The Hunt For Red October/ for the gazillionth time, I was curious to know if "magnetohyrodynamic drive" was ever actually a thing or not. I asked Google, and I got a very authoritative-sounding answer saying "yes! In fact, the US just launched a new submarine with a drive like this". When I looked into it, their AI was citing an April Fools' Day article from a few years ago.
Turns out AIs are just as gullible as humans are when it comes to seeing something on the internet and assuming it must be true.
I recently fell for a Hunt for Red October April Fool's joke as well! I've never seen the movie before so I watched this video about a rogue Soviet submarine defecting to the US from a channel that otherwise provides straight-laced commentary, and I just kept thinking "why the fuck have I never heard about this?" as I tried to look up articles about the story.
Using AI as an editor for a fiction project, and one interesting thing to note. I've had times where I've run a couple paragraphs by it, taken some of the advice it gave me, then run the edited version by a fresh instance and gotten literally the opposite advice.
Just something to keep in mind. Sometimes the advice is good, sometimes it's not, and sometimes it's a matter of particular tone and style. It's still quite useful, and I often have it as more of a "beta reader". How does character X come across in this intro scene, can you tell what is Y implying but not saying, that sort of thing.
That's my experience as well; the feedback it gives is often erratic. It remains very useful when offering up a bevy of alternative options I can pick and choose from.
Yassine - this is excellent and I have gained some great new ideas for my own use cases based on yours. I am curious, though, would you ever be willing to share some of the prompts you used to help facilitate the examples in each of the categories? I love the ideas you provided and would be especially grateful to see what prompts/chats you used as starting points to get the results you wanted/that were helpful for you.
I didn't think to share them because I assumed they would be obvious. There's no fancy prompt engineering going on here at all, chatGPT is really excellent at understanding naturalistic commands though sometimes it takes a few iterations of feedback to get it right. My favorite way to use it is via the voice input feature and just babble and infodump for 5 minutes, giving it as much useless context as possible and hoping it sorts it out. It almost always does!
I think I can understand versions of this concern, but not as an overall principle. The thesaurus aspect is just far better than spelunking through thesaurus.com, which in turn is far better than flipping through a hardcopy tome.
I came here to say the same thing. When I first learned about INDEX(MATCH, it felt like unlocking a secret level. And anyone else who "knew" (i.e. that VLOOKUP *sucked!*) was automatically in that secret club with you. They got it.
But now, XLOOKUP is just the shit. It's like the born-again version of VLOOKUP that works how it was supposed to all along.
"I’ve tried feeding AI giant chunks of my writing to see how well it can generate something from scratch but that exercise has thus far always been a disappointment. It never gets it quite right, nor does it avoid a dry and monotone delivery."
These days, if I try to be honest with myself and look for gold I might be leaving on the ground out of laziness and inertia, voice transcription is at the top of the list.
Would you recommend any specific software or model? Ideally ones that can be self-hosted?
Surprisingly the choices are still between "free and kinda bad" (Google Recorder, etc) or "cheap but awkward" (chatGPT premium) or "good but expensive" (Rev.com and the like).
Okay but what about that spoken language practice? I asked ChatGPT and it assured me it would not speak to me in a different language. I asked it whether there's an AI that can, and it failed to find me anything.
I'm baffled because I have premium and it told me it would not speak in the language I specified. Do you change a language setting, or do you simply speak to it and ask it reply in the same language?
I've written a few posts here and there over the years. I've never used AI and have no plans to do so. I kinda like picking my own words, doing my own research, thinking my own thoughts and expressing them in whatever way I choose. But then, I'm old. On the other hand, whatever I write is mine, for better or worse.
One of my favorite use cases is generating recipes for dinner. I list a bunch of ingredients I want to use up or are about to go bad and it almost always comes up with something decent. It's especially good at making very specific substitutions: "rewrite that chicken stir fry recipe to use canned chicken instead of fresh chicken breasts" for example.
Yes! I also used it to get feedback on how to optimize familiar recipes if I'm interested in streamlining the process (e.g. when should I drop in greens? how should I avoid splatter? etc).
INTERESTING
That Python thing sounds like the kind of thing you should try and monetize. A very niche problem with a solution that saves a ton of time while also vastly improving usability. Deserves to make money!
Recently, after watching /The Hunt For Red October/ for the gazillionth time, I was curious to know if "magnetohyrodynamic drive" was ever actually a thing or not. I asked Google, and I got a very authoritative-sounding answer saying "yes! In fact, the US just launched a new submarine with a drive like this". When I looked into it, their AI was citing an April Fools' Day article from a few years ago.
Turns out AIs are just as gullible as humans are when it comes to seeing something on the internet and assuming it must be true.
I recently fell for a Hunt for Red October April Fool's joke as well! I've never seen the movie before so I watched this video about a rogue Soviet submarine defecting to the US from a channel that otherwise provides straight-laced commentary, and I just kept thinking "why the fuck have I never heard about this?" as I tried to look up articles about the story.
https://www.youtube.com/watch?v=TvAueVn6Fzo
Using AI as an editor for a fiction project, and one interesting thing to note. I've had times where I've run a couple paragraphs by it, taken some of the advice it gave me, then run the edited version by a fresh instance and gotten literally the opposite advice.
Just something to keep in mind. Sometimes the advice is good, sometimes it's not, and sometimes it's a matter of particular tone and style. It's still quite useful, and I often have it as more of a "beta reader". How does character X come across in this intro scene, can you tell what is Y implying but not saying, that sort of thing.
That's my experience as well; the feedback it gives is often erratic. It remains very useful when offering up a bevy of alternative options I can pick and choose from.
Yassine - this is excellent and I have gained some great new ideas for my own use cases based on yours. I am curious, though, would you ever be willing to share some of the prompts you used to help facilitate the examples in each of the categories? I love the ideas you provided and would be especially grateful to see what prompts/chats you used as starting points to get the results you wanted/that were helpful for you.
I didn't think to share them because I assumed they would be obvious. There's no fancy prompt engineering going on here at all, chatGPT is really excellent at understanding naturalistic commands though sometimes it takes a few iterations of feedback to get it right. My favorite way to use it is via the voice input feature and just babble and infodump for 5 minutes, giving it as much useless context as possible and hoping it sorts it out. It almost always does!
The Excel formula language is, in fact, Turing-complete. No need to pooh-pooh that.
In my experience, people who look down on Excel are usually ignorant of just how much it is stuffed with advanced functionality.
I learned a new word!
I find this post quite depressing. It seems like AI is being used to do all the parts of writing that I find interesting if challenging
I think I can understand versions of this concern, but not as an overall principle. The thesaurus aspect is just far better than spelunking through thesaurus.com, which in turn is far better than flipping through a hardcopy tome.
For (almost) all use cases, XLOOKUP is the way to go.
I came here to say the same thing. When I first learned about INDEX(MATCH, it felt like unlocking a secret level. And anyone else who "knew" (i.e. that VLOOKUP *sucked!*) was automatically in that secret club with you. They got it.
But now, XLOOKUP is just the shit. It's like the born-again version of VLOOKUP that works how it was supposed to all along.
R.I.P., INDEX(MATCH. You did good.
It seems like a relatively new formula? I'm glad I don't have to worry about remembering these anymore.
"I’ve tried feeding AI giant chunks of my writing to see how well it can generate something from scratch but that exercise has thus far always been a disappointment. It never gets it quite right, nor does it avoid a dry and monotone delivery."
My experience is the exact same here.
"strangle yourself from getting your panties in a bunch"
Did ChatGPT help with this metaphor? Where are we wearing our panties?
That mixed metaphor was 100% homegrown
These days, if I try to be honest with myself and look for gold I might be leaving on the ground out of laziness and inertia, voice transcription is at the top of the list.
Would you recommend any specific software or model? Ideally ones that can be self-hosted?
Surprisingly the choices are still between "free and kinda bad" (Google Recorder, etc) or "cheap but awkward" (chatGPT premium) or "good but expensive" (Rev.com and the like).
Okay but what about that spoken language practice? I asked ChatGPT and it assured me it would not speak to me in a different language. I asked it whether there's an AI that can, and it failed to find me anything.
Premium chatGPT can speak multiple languages, it even was able to distinguish between classical Arabic, and "dareeja" Arabic from Morocco.
I'm baffled because I have premium and it told me it would not speak in the language I specified. Do you change a language setting, or do you simply speak to it and ask it reply in the same language?
I just spoke to it in the language. Tried it just now with both its speech2text and voice chat options with model 4o and it worked without any issues.