Typo from an AI. WTF? Never saw this before.

Because of the kind of work I do, I use LLMs. A lot. Hours and hours every single day since about Christmas of 22. About 15 minutes ago I asked for general information about devices known as EPAPs. I got this:

That's a great question. Expiratory Positive Airway Pressure (EPAP) devices are used primarily to treat Obstructive Sleep Apnea (OSA).

Here are some examples of FDA-cleared or approved EPAP devices:

  • Bongo Rx: This is a reusable, nasal EPAP device that is FDA-cleved for the treatment of mild to moderate Obstructive Sleep Apnea (OSA). It uses soft silicone nasal seals and works by having small valves open during inhalation and close during exhalation, directing the air through small vent holes to generate the needed pressure.
  • ULTepap: This is another reusable EPAP device that seals the nares (nostrils) with a nasal pillow interface. It is also FDA-cleared for the treatment of OSA.
  • … other information.

In the entire time I've been using LLMs I have never seen a typo like this. All kinds of other mistakes, but not a flat-out, clear, unambiguous typo in response to a one-line query "Tell me about epap devices for apnea." Wtf?

I can see where an LLM may have used a word someone was unfamiliar with and therefore the person thought it was a typo. Or the LLM may have given a correct response that someone thought was a typo, e.g., "peaked my interest" (wrong but often mistaken as right) versus "piqued my interest" (right but often mistaken as wrong because someone doesn't know what pique means). But I have never seen an actual typo before.

Leave a Reply