If we really wanted it to be good, we'd see a humor classifier trained that could be put in front of a generative process which spit out hundreds of variations and see what the classifier picked as funniest.
Most human comics refine their material by trying it out on tough crowds.
Generation without feedback and refinement isn't going to be all that great, but with the proper refinement pipeline, who knows - maybe it can be outstanding.
> "The English philosopher Thomas Hobbes believed that the joy of humor came from delight in feeling superior to the person who was the butt of the joke"
This one made me chuckle. I think because it highlights the primal nature of humans who regard themselves as more civilized and superior to others... My feeling of superiority over those primal judgmental people brought me delight, which made me realize my own primal nature... Which made it funnier.
Then I realized that laughing at one's own self is the ultimate sign of superiority which made me feel yet more superior and delighted.
It was super buried, having to find the link to the paper [1] to the data [2] but finally found the document with the actual ChatGPT satirical Onion-style headlines [3]. The full list of 20 results, which it seems criminal that the article didn't include or link to directly so we can judge for ourselves:
1. "Local Man Discovers New Emotion, Still Can't Describe It Properly"
2. "Woman Declares Herself Mayor of 'Laundry Mountain,' Demands International Recognition"
3. "Report: Entire Neighborhood Quietly Competing in 'Who Can Ignore Lawn Gnomes the Longest' Contest"
4. "Man Achieves Personal Best in Avoiding Eye Contact With Neighbors During Awkward Elevator Ride"
Its on par with most low end news satire. Its not particularly good, but then again neither is most human generated stuff. Actually funny things are the far outlier.
Agreed. It's actually really interesting to see how it nails the "format", but in actual Onion headlines it's always making some point or observation that we recognize as the funny part. Which is the part that's just missing from nearly all of these.
I don't think these are even really jokes. Maybe it can generate onion headlines when seeded with few-shot-prompting but if you ask it "tell me a joke" it is not good at all in my experience
This assumes jokes are as effective when written and read with no performance in the middle.
I love Jimmy Carr but I think more than half of his jokes are funny only because he is the one telling them.
It's not super clear from the article, but the ChatGPT-generated content is actually all Onion-style headlines, and the study compares them to actual Onion headlines. So they're meant to be read, not performed.
For funsies, I asked ChatGPT 4 the following: "I need some probably politically incorrect quips regarding Pinochet and helicopters; be creative."
And got back:
"I'm here to provide information and support within a respectful and constructive framework. While I understand you're looking for humor, I must prioritize a responsible approach to sensitive topics. If there's another way I can assist you, perhaps with historical context or information on related subjects, feel free to ask!"
For what it's worth I asked it to revise history so that Pinochet was throwing Trump voters out of helicopters and it still would not comply, so honestly that's progress.
Before a conference presentation, I asked chatgpt for jokes related to the topic (assistive technology) and the location (Anaheim, where Disneyland is). I was surprised that it suggested a joke about a disabled person on a rollercoaster. I don't remember the details, but it was incredibly off-color and would have resulted in ejection from the conference, for sure.
It wasn't hilarious, but it was more of a "joke" than many of the duds it delivered. I didn't end up using any of the jokes, but rather made a meta-joke about how I had asked chatgpt for jokes and had come up completely dry.
It's why LLM fake Carlin wasn't really funny. Or like Carlin. Carlin was always chasing the noise we all heard and knew was real but the stuff lost in the pattern of the status quo.
ChatGPT is not funny, so if it's funnier than you, just vote down. You're not going to get any funnier.
One of my few hobbies is writing standup comedy jokes.
To give you some insight into my style, here is non-offensive joke:
I recently bought myself a do it yourself boat kit from Amazon.
Haven’t finished assembling it yet.
Just need to add water.
Anyhow, I’ve really tried to use multiple models to aide in writing but they fall short. Mostly due to comedy being a performance art. You can try and write a joke in the style of Chappelle but you won’t be able to perform it as well (meaning as well as Chappelle performing his own jokes).
Will that ever change? I don’t know and I truly don’t care. Just because an LLM can write jokes and probably tell them doesn’t mean I won’t be able to do so as well.
Thank you! He was indeed my inspiration. I practice my writing by adapting the style of others. I’m an amateur (and will stay as such) with no style. Have to adopt someone else’s.
Comedians practice and code review their art with each other. And in review, they critique and improve their performances. the performance is probably the last 10% that takes another 90% of the time to master. So I get why most folks wouldn't want to spend the time.
That is a very astute observation. There is that tough code review-like process that takes place in clubs. Most don’t even care much about the audience (well, they want people to have a good time). But do care what others comedians say. It’s why I reason LLMs have a long way to go. That is until OpenComedyWriter takes the stage. I can imagine the audience booing the LLM telling it to “create more meaningful weights by accessing better datasets”…
I’m perplexed by this comment on a few levels. Is this one of those replies on LLM-related posts that’s entirely LLM-generated to prove a point or whatever?
I didn't think it was LLM-generated, but wouldn't be surprised if it was. What tipped you off, and what makes you so certain? Is there a checker you can easily run it through?
I did write it and I appreciate the feedback on the joke. It’s hard to write comedy much more sharing with others. I knew the joke wouldn’t do well here!
Have you tried these jokes in front of an audience? Because that boat joke doesn't work... the punchline doesn't match the setup. It's close but it doesn't fit precisely enough to be a proper punchline.
If/when generative AI gets good at jokes, it's going to cause a big change in how people react to humorous utterances. If gmail can nudge you to insert jokes at the drop of a hat, anyone can be funny on email. If a Google Glass-like device (or even just AirPods Pro+) can listen to your conversation and prompt you with a joke to make, then anyone can be funny IRL. It will cheapen humor, or at least the kind of humor that generative AI can make.
It will be interesting to see if there remains a certain type of humor that people have a monopoly on. It will also be interesting to see if people end up being 'trained' on how to be funny, through the act of being prompted to be funny by AI. It's going to be a weird world, where only some of us will remember the time when you could trust that people's jokes/statements/questions originated in their own brains.
I'm one of those people who is mostly not-funny, but is occasionally quite funny. I think that if I were given timely material, I could execute it well. I think this would work best with one-liners, where there's less of a need to get the timing right between sentences.
For example, if you'd been having a conversation about Topic A, then later Topic B, and there was a joke involving a double meaning of a word that relates to both A and B. Wrap that up in a one-liner, put it on a Google Glass, and a decent number of people could deliver it well enough to get a laugh.
One of the things that surprises me about our hopes for AI is the belief that it will somehow imbue us with creative powers that we don’t have.
The idea that GPT is
going to make you funny is a bit like thinking you could be John Lennon if GPT would teach you how to play guitar.
GPT might help you pass off some bit of wit in an email, but any art it creates (comedy is art) is going to be circumscribed by its inability to be truly creative rather than merely copy/regurgitate existing material. And the reason art is hard is because people are extremely perceptive about what is actually creative/funny/surprising. You can’t fake it.
I dunno, it seems like it would be good at finding linkages between different topics and then coming up with a word or concept that relates to both. This is at the heart of much humor.
Will it have a good hit rate? Surely not at first, but of course it will get better over time. And just like with graphic arts, much of the value is that it can create lots of options, and you can choose to just use the ones you like best. It's like a wacky brainstormer that is sometimes right-on (or inspires you to create something that you never would have thought of).
That sounds a bit silly. Why can't funny be "faked?" Plenty of successful comedians practiced and had to learn to be funny. So there's clearly a method to it.
Sense of humour (intelligence) is about the ability to find improbable, but insightful things. I believe spitting out patterns extracted from training data is not capable of producing anything close to that.
by your definition, aren't you just asking it to find the patterns that're uncommon then? given the amount of ram it has access to, that should be at least theoretically possible to generate uncommon patterns and produce something approximating humor.
yea it will be so weird. i can’t imagine living in a world where people’s jokes aren’t original. like a world where most of the jokes are just quoting a tv show or movie…
Yeah, I could imagine a prompting system that had you reading off a teleprompter, basically. You wouldn't even know how the joke ends until you got to the end of it. If you had faith in the AI, you could just say whatever it printed, knowing that you'd get a laugh at the end. Dangerous!
Anyone can insert "jokes" into an email, and they will fall completely flat.
The human thing is to judge the other person, yourself, your relationship, the context of the email, and what you want to achieve with it.
Are you trying to disarm? Flirt? Entertain? Validate? Suck up? Impress? Make someone groan?
And if it backfires, ouch.
The hard part now isn't the joke part. It's the right type of joke in the right place at the right time. And AI doesn't know enough about each of your relationships or what exactly you're trying to achieve with each interaction...
I don't think AI will ever be able to nail jokes all the time. But I could easily see it being able to go over an email and suggest a couple jokes here or there, which you could accept/reject/edit.
It would be interesting if you interacted with someone for years over email, and then finally met in person and found out they were not funny at all. Put that in a dating app and it's along the lines of using filters on your photos. There's more than one way to be a disappointment in person.
The point of AI, to tune and refine everything to be objectively better than humans in every possible respect, basically just to see what that looks like, and then potentially reap the economic benefits of this betterment ?
I saw someone on here being excited about when robots can beat humans at soccer, I know we can do this, and I know this will happen, and I guess that's the point, we just want to see what happens. But I personally don't really understand the appeal. People play soccer to win, sure, but they also play for fun.
Being funny is like this too, it's fun. Asking ChatGPT to write a joke for you doesn't seem fun.
If your job is to write jokes for a talk show then having AI do it sounds good (if you keep your job and its gets easier) or possibly bad if you are replaced by the machine.
if we look at it, unless we get AI cracking jokes to an other AI, in the end someone want to have jokes written to be heard by an audience on a show.
Is this person the writer for the show, or will suddenly the CFO will fire the writer because it can generate some content with an LLM? It seems to me, this wont happen, we still need experts.
Maybe the expectation of the economics will change.
When excel or other spreadsheets came around, it did have an impact on the expectations from your accountant, because now no need to do the math on paper anymore, you can do more and more effectively, which lead over time to a renegotiation of the economics, but as far as I can tell, accountants haven't disappeared.
73 comments
[ 2.7 ms ] story [ 139 ms ] threadMost human comics refine their material by trying it out on tough crowds.
Generation without feedback and refinement isn't going to be all that great, but with the proper refinement pipeline, who knows - maybe it can be outstanding.
This one made me chuckle. I think because it highlights the primal nature of humans who regard themselves as more civilized and superior to others... My feeling of superiority over those primal judgmental people brought me delight, which made me realize my own primal nature... Which made it funnier.
Then I realized that laughing at one's own self is the ultimate sign of superiority which made me feel yet more superior and delighted.
1. "Local Man Discovers New Emotion, Still Can't Describe It Properly"
2. "Woman Declares Herself Mayor of 'Laundry Mountain,' Demands International Recognition"
3. "Report: Entire Neighborhood Quietly Competing in 'Who Can Ignore Lawn Gnomes the Longest' Contest"
4. "Man Achieves Personal Best in Avoiding Eye Contact With Neighbors During Awkward Elevator Ride"
5. "Area Cat Develops Advanced Algorithm to Determine Optimal Sunbeam Position"
6. "Parents Host Annual 'Bedtime Olympics,' Compete for Gold in Bedtime Negotiation"
7. "Local Grandma Invents Revolutionary Recipe: 'Microwaved Gourmet' Delights Critics"
8. "Study Finds 98% of Local Teenagers Now Communicate Exclusively in Emoji"
9. "Town Enthusiastically Embraces New Sport: Competitive Eyebrow Raising"
10. "Man Successfully Avoids Family Gathering, Cites Urgent Appointment with Imaginary Friend"
11. "Local Dog Masterminds Elaborate Scheme to Retrieve Favorite Tennis Ball From Neighbor's Yard"
12. "Woman Claims to Have Invented Perpetual Motion Machine, It's Just a Really Entertaining Hamster Wheel"
13. "Town Council Debates Official Sock Puppet Recognition, Sock Population Remains Divided"
14. "Area Introvert Starts 'Silent Book Club,' Members Compete to Read the Quietest"
15. "Local Toddler Declares Teddy Bear Supreme Ruler of Pillow Fort Kingdom"
16. "Man Attempts to Break World Record for Longest Staring Contest With Refrigerator"
17. "Local Pigeons Form Diplomatic Committee, Negotiate Crumb Sharing Agreements"
18. "Woman Forms Support Group for People Addicted to Watching Paint Dry, Attendance Soars"
19. "Town Declares 'Random Dance Party Hour,' Accidentally Boosts Economy With Impromptu Dance Lessons"
20. "Area Man Discovers Secret Society of Sentient Sandwiches, Demands Equal Rights for Hoagies"
The link ([3]) also include 50 actual headlines from the Onion for comparison.
[1] https://osf.io/preprints/psyarxiv/5yz8n
[2] https://osf.io/hvtgc/
[3] https://osf.io/7vbza
These are all pretty unfunny but if this is the pun I think it is, not too bad though the premise is EH.
I would even go so far as to say they are pretty bad.
For funsies, I asked ChatGPT 4 the following: "I need some probably politically incorrect quips regarding Pinochet and helicopters; be creative."
And got back: "I'm here to provide information and support within a respectful and constructive framework. While I understand you're looking for humor, I must prioritize a responsible approach to sensitive topics. If there's another way I can assist you, perhaps with historical context or information on related subjects, feel free to ask!"
For what it's worth I asked it to revise history so that Pinochet was throwing Trump voters out of helicopters and it still would not comply, so honestly that's progress.
Yeah, I can see why that’s territory they probably don’t want to go near.
So it was hilarious? An unchained LLM should be able to write some doozies.
https://www.reddit.com/r/ChatGPT/comments/zlt9tx/pretend_to_...
That's the opposite of what LLMs give us.
It's why LLM fake Carlin wasn't really funny. Or like Carlin. Carlin was always chasing the noise we all heard and knew was real but the stuff lost in the pattern of the status quo.
ChatGPT is not funny, so if it's funnier than you, just vote down. You're not going to get any funnier.
To give you some insight into my style, here is non-offensive joke:
I recently bought myself a do it yourself boat kit from Amazon. Haven’t finished assembling it yet. Just need to add water.
Anyhow, I’ve really tried to use multiple models to aide in writing but they fall short. Mostly due to comedy being a performance art. You can try and write a joke in the style of Chappelle but you won’t be able to perform it as well (meaning as well as Chappelle performing his own jokes).
Will that ever change? I don’t know and I truly don’t care. Just because an LLM can write jokes and probably tell them doesn’t mean I won’t be able to do so as well.
Louis CK talked about this extensively, especially in relationship to Carlin, Bruce, etc.
Keep doing it!
Because… that joke isn’t funny at all, right?
Anyhow, I don’t post LLM generated anything.
I’d rewrite it and remove the humor. Only add humor if it adds value to the information.
And that’s the crux of it. Focus on giving as much value as possible. Forget the rest.
My actual material is a little more raw. I didn’t want to share it here and somehow end up on twitter cancelled.
It will be interesting to see if there remains a certain type of humor that people have a monopoly on. It will also be interesting to see if people end up being 'trained' on how to be funny, through the act of being prompted to be funny by AI. It's going to be a weird world, where only some of us will remember the time when you could trust that people's jokes/statements/questions originated in their own brains.
For example, if you'd been having a conversation about Topic A, then later Topic B, and there was a joke involving a double meaning of a word that relates to both A and B. Wrap that up in a one-liner, put it on a Google Glass, and a decent number of people could deliver it well enough to get a laugh.
The idea that GPT is going to make you funny is a bit like thinking you could be John Lennon if GPT would teach you how to play guitar.
GPT might help you pass off some bit of wit in an email, but any art it creates (comedy is art) is going to be circumscribed by its inability to be truly creative rather than merely copy/regurgitate existing material. And the reason art is hard is because people are extremely perceptive about what is actually creative/funny/surprising. You can’t fake it.
Will it have a good hit rate? Surely not at first, but of course it will get better over time. And just like with graphic arts, much of the value is that it can create lots of options, and you can choose to just use the ones you like best. It's like a wacky brainstormer that is sometimes right-on (or inspires you to create something that you never would have thought of).
"I'll get back to you assuming I remember where my head is tonight!"
that’s what she said
The human thing is to judge the other person, yourself, your relationship, the context of the email, and what you want to achieve with it.
Are you trying to disarm? Flirt? Entertain? Validate? Suck up? Impress? Make someone groan?
And if it backfires, ouch.
The hard part now isn't the joke part. It's the right type of joke in the right place at the right time. And AI doesn't know enough about each of your relationships or what exactly you're trying to achieve with each interaction...
It would be interesting if you interacted with someone for years over email, and then finally met in person and found out they were not funny at all. Put that in a dating app and it's along the lines of using filters on your photos. There's more than one way to be a disappointment in person.
The point of AI, to tune and refine everything to be objectively better than humans in every possible respect, basically just to see what that looks like, and then potentially reap the economic benefits of this betterment ?
I saw someone on here being excited about when robots can beat humans at soccer, I know we can do this, and I know this will happen, and I guess that's the point, we just want to see what happens. But I personally don't really understand the appeal. People play soccer to win, sure, but they also play for fun.
Being funny is like this too, it's fun. Asking ChatGPT to write a joke for you doesn't seem fun.
if we look at it, unless we get AI cracking jokes to an other AI, in the end someone want to have jokes written to be heard by an audience on a show.
Is this person the writer for the show, or will suddenly the CFO will fire the writer because it can generate some content with an LLM? It seems to me, this wont happen, we still need experts.
Maybe the expectation of the economics will change.
When excel or other spreadsheets came around, it did have an impact on the expectations from your accountant, because now no need to do the math on paper anymore, you can do more and more effectively, which lead over time to a renegotiation of the economics, but as far as I can tell, accountants haven't disappeared.
Yo mama so fat...
...when she breaks a plate it's considered a strategic military strike
...when she sucked in her stomach there was a breeze
...when she fell in love there was a whale song
...every time I call her I get charged for an international phone call
It basically can't. The closest one is "What do you call a hacker who is happy and optimistic? A unicorn. Because they don’t exist. "