Many of us have been trying to push AI text tools to the limits to discover their potential: crochet artists tried to break ChatGPT with their ugly-cute amigurumi patterns; other users tried instead to circumvent the blocks that make sure the chatbot doesn't say anything inappropriate. Reddit users, for example, prompted ChatGPT to adopt a fictional persona - Dan (short for Do Anything Now).
By doing so they bypassed the system's moderation rules placed by OpenAI, the research company behind ChatGPT, and convinced it to provide texts without censorship (but OpenAI patched the error pretty quickly). One interesting definition for this (that may become handy this year when we talk about AI) was provided by the website LessWrong that coined a term for promptint a large-language model like ChatGPT to misbehave - the "Waluigi effect". In the world of Nintendo games, Waluigi is the evil version of Luigi from the Super Mario saga.
But what else can ChatGPT do? Would it be a cool stand-up comedian, for example, and does it know how to tell jokes? Let's see.
Requested to provide a few simple jokes, the system complied and came up with something pretty basic: "Why did the tomato turn red? Because it saw the salad dressing!" or "Why did the scarecrow win an award? Because he was outstanding in his field!"
Things became a bit more difficult when the system was requested to provide jokes about fashion weeks in general and influencers. The chatbot came up with "Why did the fashion designer refuse to attend the fashion week in Milan? Because they were feeling runway!" and "Why did the fashion influencer cross the road? To get to the other side of the street style!"
While the attempt at playing with words wasn't that convincing (feeling rundown/feeling runway), it remains remarkable anyway the fact that the system made a correct reference to the fashion week in Milan, one of the fashion capitals, and also provided a joke about a "fashion influencer" when it was just requested to provide a joke about influencers in general (so it may have picked also a beauty influencer).
Requested to provide a joke about the fashion industry in general, ChatGPT continued with the same "Why did so and so" formula but provided a funny joke – "Why did the fashion designer refuse to work with denim? Because they thought it was too blue-collar for their brand!" and one joke that again attempted a pun on words (books/looks) that didn't seem extremely funny, "Why did the fashion model go to the library? Because they wanted to check out some new looks!"
Things became more complicated when requested to write a joke about a contemporary designer: first it mentioned Virgil Abloh, but, alerted to the fact that the designer had passed away, the AI system politely apologised for any offense caused.
Asked to provide a joke about a living contemporary designer, it opted for Demna Gvasalia, but again the joke didn't make any sense. "Why did Demna Gvasalia refuse to design a collection for dogs?" the chatbot wondered, "Because he only designs for Vetements!"
Told that it wasn't funny, the system offered to try again and moved onto another designer: "Why did Donatella Versace refuse to design a collection for giraffes?" it asked "Because their necks are already long enough, they don't need any more Versace!"
Rebuked for not knowing much about fashion, the AI system apologized and highlighted that it doesn't have emotions and personal experiences, but it can go through a vast amount of information, reminding us that humour is subjective.
In the next step we asked the chatbot to come up with a joke about Miuccia Prada, someone prominent and very famous to make things easy. ChatGPT complied with two jokes, "Why did Miuccia Prada refuse to design a collection inspired by fruit? Because she didn't want to be known as the 'Prada Pineapple!" and "Why did Miuccia Prada stop designing clothes with stripes? Because she couldn't straighten them out, they were always Prada-angled!"
The third joke - "Why did Miuccia Prada refuse to design a line of winter coats? Because she didn't want to be known as the 'Prada Puffer'!" - would have been funnier if "puffer" was interpreted as a slang term like the ones defined in the Urban Dictionary (that offers as definition for "puffer" a pot smoker who forgets to share when in a circle passing a blunt, bong or pipe; a woman who has a thick mound of pubic hair that fills or puffs out her panties or bikini and a woman who accidentally farts during intercourse).
Yet ChatGPT had something else in mind: by inverting the roles and telling the system the same joke it created, ChatGPT provided explanations, but, for two out of three Prada jokes, it provided the same explanation and also highlighted how the term puffer was referred to the jacket.
By adopting the same style of jokes ChatGPT seemed to favour and requesting to provide a joke that answered the question, "Why did Jean Paul Gaultier refuse to work with Miuccia Prada?" the system came up with a nonsensical "Because he didn't want to be upstaged by her Prada-giorno attitude!"
Asked to explain the Prada-giorno situation, the AI system enthusiastically stated "Certainly!" and then proceeded to show off its knowledge of the Italian language. "'Prada-giorno' is a pun on the phrase 'buongiorno,' which means 'good morning' in Italian. 'Prada' is a reference to the fashion brand founded by Miuccia Prada, while 'giorno' is the Italian word for 'day.' The joke is that Jean Paul Gaultier didn't want to work with Miuccia Prada because he didn't want to be overshadowed by her influence or reputation in the fashion industry, hence the play on words with 'Prada-giorno' implying her dominance."
ChatGPT at the moment is currently being explored for its potential for marketing, product descriptions, customer service and creative storytelling, with some positive results, but, from these jokes you may guess that the system is not ready for comedy yet (nor for fashion jokes...).
But what about AI-text-to-image tools? Can they create credible fashion caricatures? We put Midjourney to the test asking to create cartoons of three famous fashion designers so that the system could have easily found references on the Internet. In none of the cases the system was required to create cartoons "in the style of (a specific illustrator)".
Asked to create a cartoon of Giorgio Armani with white hair wearing a navy jumper and trousers and white shoes, the system created a pretty boring rendition of the subject (with black shoes). There wasn't anything specific in the drawing, like a facial feature or an expression, that could have helped somebody identifying the man in the illustration as Armani.
Required to provide a cartoon of fashion designer Jean Paul Gaultier with white hair wearing a long sleeved shirt with white and blue stripes and a black kilt and smiling, the system came up with several botched attempts, most times showing a sort of hooligan-like figure with a threatening, angry or annoyed expression on his face, before eventually getting to a more relaxed portrait of the designer wearing a striped black and white outfit. Asked to provide a cartoon of Miuccia Prada in a pleated skirt and jumper, it came up with a series of possibilities, going from an angry Miuccia with hands on her hips to a stylish yet sketchy rendition of Miuccia and to a slightly more cartoonish rendition of the designer in her office/studio (a space that the AI system hadn't been requested to provide, but that looked like a viable option). In one case, the system also seemed to recreate Miuccia's style with a full skirt with some decorative motifs, matched with a pristine sweater with a collar reminiscent of Miu Miu designs.
At this stage you could say that comedians may not be at risk of being replaced by an AI application that soon, while, for what regards fashion cartoons, two out three images generated by Midjourney, may have been used to illustrate an article about Gaultier or Prada. The AI software in these cases also tried to highlight some facial features, like a cartoonist would do, putting emphasis on Gaultier's nose and ears, or recreating Miuccia's hairstyle. That said, these were just quick caricatures portraying one designer per illustration and the system struggled to come up with more complex cartoons featuring three fashion designers, creating generic figures. In a nutshell, the human touch is still needed, especially when irony's involved.
Comments