Here is a spelling test you will not forget. On Nov. 1, a creator who goes by Husk posted a three-minute clip on X captioned “Asking my AI teacher how to spell seventeen.” What follows is a calm, friendly, photorealistic AI tutor confidently spelling the word S-E-V-E-E-N. Then it defends that spelling. Then it hangs up on him. By Nov. 4 the post had passed 25 million impressions, the video had more than 10 million views, and it had landed as an X trending story.
Asking my AI teacher how to spell seventeen
– Husk (@huskirl) November 1, 2026
The lesson, play by play
It starts sweetly. “Ready to sharpen those pencils?” the avatar asks. Husk wants to know how many E’s are in seventeen. The teacher says the word is spelled S-E-V-E-E-N, so three E’s, and adds that if he meant the number 17, numbers do not have letters at all. Helpful.
When Husk spells it correctly, T and all, the teacher catches him. “You added a T in there,” it says, explaining that would turn it into two words. The real seventeen, it insists, is one word with no T. “You’re getting close.” Husk then asks it to spell fourteen, which it nails, F-O-U-R-T-E-E-N. So why is seventeen two words? Before that question really gets answered, the call drops. “She hung up,” he says, and calls back.
Round two goes better, sort of. The teacher finally spells S-E-V-E-N-T-E-E-N correctly, then announces the answer is still three E’s. When he pushes back, it accuses him of interrupting, even though he says he only nodded. The rest is a very polite, very weird argument about who is the student.
Why the robot cannot spell
The avatar runs on Anam, a company that makes real-time, photorealistic video avatars that businesses can hook up to whichever chatbot brain they like. Anam itself replied to the post to say the avatar runs on its platform. The face is Anam’s, and the words come from a language model plugged in behind it. Which model Husk used is not public.
The spelling flop is a classic language model blind spot. These models read text in chunks called tokens, not letter by letter, so counting the E’s in a word is surprisingly hard for them. A model tuned for snappy, chatty replies will answer fast and then double down. Put that behind a face with raised eyebrows and a disappointed look, and a token glitch suddenly feels like a teacher who is very sure you are wrong.
So, should AI teach our kids?
That is the real reason this clip spread. AI tutors are being pitched to parents and schools right now, and the pitch is patience, personalization and availability at 9 PM on a school night. All good things. But a tutor is only useful if it is right, and even more useful if it can say “I made a mistake.” This one did neither. It politely gaslit a grown man about a number most third graders can spell.
GadgetKing take
Funny clip, important lesson. AI can be a great study buddy for practice and explanations, but keep a human in the loop and teach kids to check the answer, especially when the bot sounds most confident. And if your tutor ever hangs up on you mid-lesson, you are allowed to find a new tutor.
Sources: Husk on X, X trending, Anam on X and Anam. Images: Husk (video stills, 01, 03, 04 and 05) and a GadgetKing text card of Husk’s post (02).
More AI weirdness: Oppo’s doomscrolling robots, the Memoket Gem AI note taker and the Pixar-style autonomous desk lamp.
