An image of descending colored lines on a black background—like a faulty digital TV screen or, if you listen to some posters, the brain of an artificial intelligence—is going viral after people discovered that whenever you input it into ChatGPT it crashes the program.
“Whatever you do, don’t show ChatGPT this image,” intoned @brandon_xyzw mysteriously on Wednesday.
“Go ahead, try it. See how that works out for you,” they wrote in a follow-up post.
Plenty of users did just that, all getting the same result. When you feed the image into ChatGPT premium, it knocks the machine out, with all output stopping and a “Hmm…something seems to have gone wrong” response in red. That’s a common response on ChatGPT when the machine attempts an answer that overtaxes its capacity, but it left posters wondering just what it was about the image that could possibly do that.
The Daily Dot confirmed that uploading the file does crash ChatGPT.
Another user tried feeding in an image of the error message that included the photo and that ended up fixing it, with ChatGPT describing the image as “an abstract pattern of vertical lines with varying colors.”
That video looks a little like the image but it’s hard to tell if it’s an exact match.
Posters riffed on what the image might mean, with one joking that the image is “the file the[y] upload to the mother ship in Independence Day,” referencing a computer virus that takes out an invading alien ship’s defenses in the 1996 Roland Emmerich blockbuster.
“this will make the coming terminator war much shorter,” quipped @ryanjanssen.
Other users tried feeding the image into other LLM assistants like Google’s Bard and Microsoft’s Copilot, where they found that the chat assistants could handle the image without much trouble.
“Copilot with Bing is responding fine,” reported @Himanshu_nitrr, who posted the description Copilot gave of the image, which compared it to “digital noise or a glitch effect.”
“Bard seems to be fine,” added @dictator008, posting its description of the image.
Other posters had some more mundane possible explanations for the quirk.
“I bet the guys at OpenAI are using a naïve text search on words like ‘error’ in the response by the vision API. This image of a ‘digital glitch error’ then gets a false positive on their error flag function,” proposed @Lispi_Frank.
OpenAI didn’t respond to questions asking how the glitch might be explained.