Excellent info--I learned a lot. I asked Chatgpt the same question about cat jokes and it gave me the SAME replies that it di to you. So it just searches the web for apparently the most used. I re-directed it, asking it to make its own up and be sure they are not found elsewhere. It actually ttok some time to respond then gave: (not very funny)
1. My cat stared at the wall for three hours.
I finally joined her.
We’re both waiting to see who blinks first: the ghost or the landlord.
2. I tried to discipline my cat for knocking over a plant.
She blinked slowly and whispered,
"You live in my terrarium now."
3. I bought my cat a self-cleaning litter box.
Now she just sits in front of it all day,
waiting for it to finish evolving.
4. My cat has a new hobby: judging my life choices.
Every time I open the fridge after midnight,
she whispers, "Again?"
5. I asked my cat if she loved me.
She knocked my glass off the table and said,
"Define 'love'… and hurry. The dog’s watching."
6. My cat took my spot on the couch.
I asked politely for it back.
She responded by licking herself.
I think that was a “no.”
7. I bought a toy mouse that squeaks when touched.
My cat gifted it to me on my pillow at 3 a.m.
Message received: “You squeak too much.”
8. I tried to meditate with my cat.
She purred in harmony for a moment, then bit my toe.
Apparently I reached inner peace before she did.
9. My cat wrote a memoir:
“From Litter to Legend: One Nap at a Time.”
The audiobook is just purring.
10. My cat knocked over a glass, shredded a curtain, and erased my document.
Yeah, not top-shelf material. I don't even think ChatGPT actively searches the web for the jokes - it just defers to its training data as most LLMs do. Unless you explicitly ask it to find jokes on the web, that is.
What I find works best (as I also wrote in the post) is giving LLMs specific constraints. So instead of asking for a joke about "cats," ask for a joke about "cats with lightsabers teaching a history class about the Battle of Waterloo" or whatever. The resulting jokes aren't guaranteed to be funny, but they'll definitely be unique, because the LLM doesn't have an existing library of such niche material to pull from, so it's forced to create something novel.
One thing I noticed is that gpt can deliver a good roast, but it always ends it with something conciliatory.
Another thing you can do is ask it to mimic styles of known comedians. I once spent lots of time massaging it to make a good comedy routine about how nobody really likes turkey for thanksgiving anyway because it’s so dry, and I think the Jim Gaffigan version turned out the best. Still it always ended with, “but all joking aside, let’s just enjoy our turkey and be thankful.”
Good observation, and yes - that's what I've noticed as well. That's back to the whole built-in politeness aspect. And I also find that giving AI a reference (comedian, comedic technique, etc.) makes the results better. I actually ask ChatGPT to use Scott Dikkers's funny filters to come up with the jokes, which often gets better outcomes!
“The Funny Spectrum”, by Daniel Nest coming soon. I’ll buy a copy of that! If you are Justin, watch “In Time” which is both parts brilliant for interesting plot and visuals but also comically bad acting from your doppelgänger and Amanda Seyfried
The Funny Spectrum, eh? I dig the title, let's get cracking on the book itself. Then again, we probably only need the title, the cover, and some hype to sell the first 10,000 copies. Let's go!
As for "In Time"...*checks Rotten Tomatoes and sees the 37% score* - uh....I'll check it out when I'm dead.
You’re either funny or you’re not. Top shelf skill. Sure you can get funnier with practice but you can’t make someone funny. That Onion 11 types of comedy article - genius - has a pic of Will Ferrell and that is my point. Btw Daniel I see a bit of Will in your pic; ChatGPT roasting you took me there. So if most of the content out there is not funny it tracks that an LLM wouldn’t be funny most of the time or at least not without a lot of prompting. I wonder if you trained a model on Will Ferrell and his ilk it could crack some of the harder comedy types? That’s the robot friend I want
I prefer to see "funny" as a spectrum, also in terms of the types of humor people enjoy. Some people might not be funny in the mainstream sense but have quirky absurdist humor. Others might generally be serious but come up with the occasional deadpan comment.
I believe most people can be funny and that you can certainly learn the process for crafting and structuring jokes in a deliberate way. That's the premise of Scott Dikkers' "How To Write Funny" trilogy of books, and they're a great read. (The 11 Funny Filters come from there.)
And then, of course, there are naturally gifted comedy geniuses like yours humbly.
I don't know about "Will Ferrel," but I've been called a fusion of Justin Timberlake and Derren Brown (during his goatee years) based on that and similar images.
"Don't be so quick to....pick a card" (*set to the tune of "Rock Your Body"*)
I'm pretty sure you can already train or prompt LLMs on a comedian's body of work and have them mimic it with some success. So your funny robot friend might be closer than you think!
"banana frisbee donkey fart burgers": new band name I call it!
Also, we kind of wrote about the same things again on the same day, although I was talking about divergent thinking in human minds... but pretty damn similar (again).
Im sorry to admit this. I laughed like hell at the donkey frisbee banana thing. Do I need help?
You will need help passing the Turing Test if this keeps happening, that's for sure! Banana!
Excellent info--I learned a lot. I asked Chatgpt the same question about cat jokes and it gave me the SAME replies that it di to you. So it just searches the web for apparently the most used. I re-directed it, asking it to make its own up and be sure they are not found elsewhere. It actually ttok some time to respond then gave: (not very funny)
1. My cat stared at the wall for three hours.
I finally joined her.
We’re both waiting to see who blinks first: the ghost or the landlord.
2. I tried to discipline my cat for knocking over a plant.
She blinked slowly and whispered,
"You live in my terrarium now."
3. I bought my cat a self-cleaning litter box.
Now she just sits in front of it all day,
waiting for it to finish evolving.
4. My cat has a new hobby: judging my life choices.
Every time I open the fridge after midnight,
she whispers, "Again?"
5. I asked my cat if she loved me.
She knocked my glass off the table and said,
"Define 'love'… and hurry. The dog’s watching."
6. My cat took my spot on the couch.
I asked politely for it back.
She responded by licking herself.
I think that was a “no.”
7. I bought a toy mouse that squeaks when touched.
My cat gifted it to me on my pillow at 3 a.m.
Message received: “You squeak too much.”
8. I tried to meditate with my cat.
She purred in harmony for a moment, then bit my toe.
Apparently I reached inner peace before she did.
9. My cat wrote a memoir:
“From Litter to Legend: One Nap at a Time.”
The audiobook is just purring.
10. My cat knocked over a glass, shredded a curtain, and erased my document.
Then sat in my lap like nothing happened.
I think I live with a tiny, elegant supervillain.
Yeah, not top-shelf material. I don't even think ChatGPT actively searches the web for the jokes - it just defers to its training data as most LLMs do. Unless you explicitly ask it to find jokes on the web, that is.
What I find works best (as I also wrote in the post) is giving LLMs specific constraints. So instead of asking for a joke about "cats," ask for a joke about "cats with lightsabers teaching a history class about the Battle of Waterloo" or whatever. The resulting jokes aren't guaranteed to be funny, but they'll definitely be unique, because the LLM doesn't have an existing library of such niche material to pull from, so it's forced to create something novel.
This answers so many questions.
One thing I noticed is that gpt can deliver a good roast, but it always ends it with something conciliatory.
Another thing you can do is ask it to mimic styles of known comedians. I once spent lots of time massaging it to make a good comedy routine about how nobody really likes turkey for thanksgiving anyway because it’s so dry, and I think the Jim Gaffigan version turned out the best. Still it always ended with, “but all joking aside, let’s just enjoy our turkey and be thankful.”
Good observation, and yes - that's what I've noticed as well. That's back to the whole built-in politeness aspect. And I also find that giving AI a reference (comedian, comedic technique, etc.) makes the results better. I actually ask ChatGPT to use Scott Dikkers's funny filters to come up with the jokes, which often gets better outcomes!
“The Funny Spectrum”, by Daniel Nest coming soon. I’ll buy a copy of that! If you are Justin, watch “In Time” which is both parts brilliant for interesting plot and visuals but also comically bad acting from your doppelgänger and Amanda Seyfried
The Funny Spectrum, eh? I dig the title, let's get cracking on the book itself. Then again, we probably only need the title, the cover, and some hype to sell the first 10,000 copies. Let's go!
As for "In Time"...*checks Rotten Tomatoes and sees the 37% score* - uh....I'll check it out when I'm dead.
You’re either funny or you’re not. Top shelf skill. Sure you can get funnier with practice but you can’t make someone funny. That Onion 11 types of comedy article - genius - has a pic of Will Ferrell and that is my point. Btw Daniel I see a bit of Will in your pic; ChatGPT roasting you took me there. So if most of the content out there is not funny it tracks that an LLM wouldn’t be funny most of the time or at least not without a lot of prompting. I wonder if you trained a model on Will Ferrell and his ilk it could crack some of the harder comedy types? That’s the robot friend I want
I prefer to see "funny" as a spectrum, also in terms of the types of humor people enjoy. Some people might not be funny in the mainstream sense but have quirky absurdist humor. Others might generally be serious but come up with the occasional deadpan comment.
I believe most people can be funny and that you can certainly learn the process for crafting and structuring jokes in a deliberate way. That's the premise of Scott Dikkers' "How To Write Funny" trilogy of books, and they're a great read. (The 11 Funny Filters come from there.)
And then, of course, there are naturally gifted comedy geniuses like yours humbly.
I don't know about "Will Ferrel," but I've been called a fusion of Justin Timberlake and Derren Brown (during his goatee years) based on that and similar images.
"Don't be so quick to....pick a card" (*set to the tune of "Rock Your Body"*)
I'm pretty sure you can already train or prompt LLMs on a comedian's body of work and have them mimic it with some success. So your funny robot friend might be closer than you think!
"banana frisbee donkey fart burgers": new band name I call it!
Also, we kind of wrote about the same things again on the same day, although I was talking about divergent thinking in human minds... but pretty damn similar (again).
You can't call your band that, as I'm in the process of officially changing my name to exactly that. "Sir Banana Frisbee Donkey Fart Burgers, Esq."
To be fair, my post is from yesterday, but I'll allow your claim of near-coincidence...
Time is relative, though. Maybe my piece was actually first and it just looks the other way to you.
Your mom---and this is 100% true---is (a) relative.
We are all relatives of somebody, right?
Don't drag me into this!
I mean like we all came from the Big Bang or whatever. We are one.
Cartoons very enjoyable. Analysis very informative.
Comment very appreciated. Positive feedback very encouraging.