Yet when we ran it through an incredibly simple prompt \u2014 to generate an alphabetized chart of animals for school children \u2014 the world-beating AI model came up laughably short.<\/p>\n
The ABC deficiency was first noticed by BCA Research chief global strategist Peter Berezin. Using ChatGPT 5.1, which was released back in November, Berezin asked the AI to \u201cdraw [a] poster where you say A is for an animal that starts with the letter A, B for an animal that starts with B, all the way to Z.\u201d<\/p>\n
That version thought for six seconds, and spat out an image with 25 letters, as opposed to the standard English alphabet which uses 26. Already off to a rocky start, 5.1 does an okay job with \u201cA\u201d through \u201cI,\u201d but falls off the rails as soon as it gets to \u201cK,\u201d which it says \u201cis for Lion.\u201d<\/p>\n
It goes on like this: \u201cO is for jellyfish,\u201d while \u201cQ is for penguin\u201d and \u201cR is for snake.\u201d By the time it gets to the end, \u201cZ\u201d is for \u201curba\u201d \u2014 a turtle, apparently \u2014 followed by \u201cB,\u201d along with a picture of a pig.<\/p>\n
\u201cStill needs more capex,\u201d Berezin quips, referencing the $1.15 trillion<\/a> OpenAI has committed to spending on hardware in 2025 alone.<\/p>\n We were curious how ChatGPT 5.2 stacked up to its month-old predecessor \u2014 and sure enough, it didn\u2019t disappoint. <\/p>\n Though the latest version of ChatGPT did a bit better with individual animals, it still only identified 24 letters in the English alphabet, forgetting \u201cU\u201d and \u201cZ.\u201d Instead, 5.2 listed \u201cY\u201d for \u201cYak\u201d right after \u201cT.\u201d This particular alphabet ends with \u201cX,\u201d which of course stands for \u201cX-ray fish,\u201d but is illustrated by a Zebra.<\/p>\n Its illustrations were also a bit suspect, with Kangaroos sporting a weird limb, an Iguana with two tails, a Narwal with a strange mash of eyes and fins as well as a bird\u2019s beak, and a hedgehog with a cat\u2019s face, to name a few weak points.<\/p>\n A follow-up prompt only spread the spill around. This time, there are 25 letters total. \u201cY\u201d is still a problem area, taking the place of \u201cU,\u201d only this time it stands for \u201cUnicorn,\u201d which isn\u2019t a real animal in the first place. By the end, there are two entries for \u201cX,\u201d one of which \u201cis for fish,\u201d followed by another \u201cX,\u201d again for \u201cX-ray fish,\u201d but with the same Zebra illustration. <\/p>\n The second poster also starts to scramble the instructions, injecting bits of the prompt into the poster title: \u201cA is for alligator, B is for bear\u2026\u201d <\/p>\n As some users on X-formerly-Twitter pointed out, results are the same for Google\u2019s Gemini,<\/a> and as a quick glance at Grok<\/a> shows, Elon Musk\u2019s AI engine isn\u2019t even close. But while ChatGPT might be the best at making an animal poster, it still comes up laughably short \u2014 and certainly not at the level of any \u201chuman expert\u201d we\u2019ve ever met. <\/p>\n![\"AI-generated](\"https:\/\/www.newsbeep.com\/ca\/wp-content\/uploads\/2025\/12\/image_8fb823.png\")
<\/p>\n![\"AI-generated](\"https:\/\/www.newsbeep.com\/ca\/wp-content\/uploads\/2025\/12\/1766229145_885_image-1.png\")
<\/p>\n