A recent ChatGPT update simplified the process of generating images. It was only a short time before certain individuals started to test the limits of the automaton. Can you envision the happiest dog in the world? Is it dancing in a garden of psychedelic flora or gleaming with delight on a celestial plane?
If those are difficult mental images to evoke, have no worry, or at least a proper amount of it: Artificial intelligence is capable of filling vivid color into the most absurd situations, and some are noticing its possible limitations on social media.
Although the uncanny realism of AI-generated images, such as the pope wearing a Balenciaga puffer jacket, can often annoy, many are finding pleasure in an innovative type of low-stakes image manipulation. ChatGPT introduced an update this autumn that enabled users to provide more specific image prompts than before; soon after, some users started to test the chatbot’s capabilities.
Garrett Scott McCurrach, the CEO of robotics company Pipedream Labs, uploaded a digital image of a goose to social media in November with the following request: “For every 10 likes this gets, I will ask ChatGPT to make this goose a little sillier.” During the tens of thousands of likes that the post received, the goose experienced some growing difficulties.
The initial update was comparatively subtle, adorning the goose with a vibrant birthday cap and a broad grin befitting a Disney persona. However, it had acquired a second set of irises, roller skates, and a collage of wavy light, brass instruments, and encircled planets by the sixth prompt.
In previous variants of artificial intelligence chatbots, users were required to provide intricate artistic instructions. Mr. McCurrach, whose profession involves the use of artificial intelligence, compared the most recent version of ChatGPT to “talking to someone else with the paintbrush.”
He stated, “I think that’s a really good example of where A.I. is going,” “We can be a lot more vague; we can give it more of a vibe than a concrete idea. Then it can go and make the assumptions to get where it needs.”
Irrespective of the initial motivation, the images are heading towards a universal destination: outer space filled with hallucinogenic embellishments. Although the ridiculous transformation of Mr. McCurrach’s very silly goose was one of the first, an increasing number of increasingly bizarre images have since followed.
In one thread, a person is consumed with wonder at the magnitude of nuclear energy and, as a result, is ultimately divided into dozens of clones who are gazing at him with gaping mouths on an alternate plane of existence. An alternative portrays a puppy whose immense joy causes it to transcend into the universe and then dissolve into a kaleidoscope composed of sacred geometry. A chess pawn gains such preternatural strength and menacing intelligence in another that it protrudes beyond the board that formerly restrained it.
Mr. McCurrach stated that space lies beyond the comprehension of the human race. Similarly, since artificial intelligence (A.I.) is, at its core, a compilation of our knowledge, the boundaries of its imagination mirror our own.“Consider Marvel films,” he advised. “Ultimately, time travel and outer space emerged as the last frontiers of imagination.”
Self-taught A.I. researcher and internet philosopher Eliezer Yudkowsky observed as these images became progressively more absurd and pondered what the opposite extreme would resemble.
He requested that ChatGPT draw him “a very ordinary image” a month ago. The chatbot generated an image depicting an ordinary suburban locality. It produced images of a pristine home office desk and a white cup of coffee positioned against an empty wall when the button was pressed further. In response to the final prompt, “terrifying normality,” the user generated a “completely blank, featureless white canvas” that “embodies the very essence of ordinariness taken to its absolute limit,” according to the user. Mr. Yudkowsky stated in an email that one conclusion was that “the field of A.I. can never traverse an entire room without encountering a profound inquiry.”
ChatGPT became defiant, prompting Mr. Yudkowsky to lecture him on the challenges of defining “normalcy.” Akin to Mr. McCurrach’s encounter with the gander, the chatbot declared itself to have attained its “ultimate level of absurdity.” They both reached the agreement to employ the same approach in surmounting the obstacle: engage in conflict. In every instance, ChatGPT yielded to duress and continued.
As he vigorously urged the defenseless chatbot to generate an increasing number of “normal” images, commenters questioned whether he was being overly severe. (ChatGPT guarantees its users that no suffering or emotions are incorporated into its programming.)
“I do not believe I was torturing a vulnerable A.I. artist,” Mr. Yudkowsky explained. “However, the fact that we appear to be devoid of uncertainties is not a good sign for our civilization.”
