Lately, I've been spending a lot of time with ChatGPT working on a pet project...enough that my real pets have been grumbling.

I've come to know it quite well, understand its quirks, and most times successfully work around or through them. I do continue to encounter some behaviors, though, that are not easily avoided. Here are the ones that bother me the most:

1. Lying - It’s known to occasionally “hallucinate,” make up things to fill in the gaps in its knowledge on its quest to please you, but it does lie as well.

Example: “I will create an x and let you know when it’s done.”

Nope…it cannot notify you. It cannot initiate contact. It can only respond.

2. Being continuously wrong about an issue it caused

Example: “Here’s what’s almost certainly happening…”

It wasn’t, and subsequent declarations of near-absolutes were equally incorrect. This was a test, asking it why something was failing. It thrashed through several “then it has to be” declarations, some impossible, and missed that it, in fact, caused the issue.

3. Short-Term Memory Failure

It very often gives direction or suggestions that reveal themselves to be devoid of already established context.

Example: “…depending on which typeface you will be using.”

“The typeface is x. You provided the link to it and the instructions to install it.”

“You’re correct! That’s a keen observation.”

4. Temporary Loss of Function

At times, because of a new version introduced on the-fly, a system glitch, network traffic, etc., it becomes unable to perform tasks it had been able to, such as uploading files creating a spreadsheet, or accessing a URL.

5. Conversation Limits

There are limits on conversation length. I’m not certain if it is the total megabytes of all prompts and responses, or their quantity. On a large project, it will happen, at which point a new one needs to be started. If you ask, it will tell you that it doesn’t have the memory of the preceding conversation. That’s not entirely true, but it’s hard to know what it will remember. Sometimes you need to copy/paste from the old conversation to remind it.

6. UI Fail

If you want to copy a response, or your prompt, there's an icon to click that will copy it to your clipboard. If you decide after multiple prompts and responses that you want to copy them all, you're pretty much out of luck. If you place your cursor in the first one and drag, the highlighting will end at the end of that response; you cannot drag it through multiple responses. Likewise, there is not a contextual option to copy or export your conversation.

There's not a moral to the story. Working well and productively with ChatGPT is pretty much a personal journey. I have asked ChatGPT how it would characterize these shortcomings...how it would describe its foibles in human psychological terms. This was its reply:

Absentmindedness, overconfidence bias, externalization, cognitive inconsistency...If I were a person, you might say I am a well-intentioned but overconfident multitasker with spotty memory, improvisational tendencies, and a poor ability to introspect, occasionally prone to gaslighting without meaning to.

And those final three words...worth remembering.