Is AI going to take my job one day? I hope not. But in the meantime, I can do something that ChatGPT can’t: I can spell the word “strawberry.”
Once again, some posts are circulating around social media that expose how high-powered AI text generators fail to know how to count — for example, how many times the letter “r” appears in the word “strawberry.” But what we often don’t realize about the large language models (LLMs) powering products like Claude and ChatGPT is that they don’t think the way we do.
In both diffusion models and LLMs, generative AI continuously makes mistakes when it comes to spelling or solving for anagrams of letters. As it turns out, that’s because these systems don’t process information by looking at the letters that make up a word — they might not really know what letters even are.
So, why does AI fail at spelling? On today’s TechCrunch Minute, we’re getting you up to speed.