They do. I probably should have stated that I’m paraphrasing here. My questions were a bit more than just keywords. When I started using ChatGPT I wrote crafted prompts and I find it works well with “broken” English/keywords like I’ve done here. Anyways ChatGPT does answer these questions just wrong. It makes up APIs.
For example ask ChatGPT how to get the computer name in Powershell. The cmdlet it reccomends doesn’t exist.
When I ask ChatGPT to enumerate the displays connected to my graphics card it uses the correct cmdlet but then when I ask it to write a program to get the EDId data it makes up an API that doesn’t exist. When confronted it says you’re right that doesn’t exist.
In the case of JupyterLab it confuses Command Pallet with Command Registry and again says that there are functions called enable() and disable() which is incorrect.
The worse was when I was trying to ask about some boilerplate image processing stuff. I wanted to do a least squares fit and it completely made up APIs or recommended used libraries 10 years out of date.
I wish I could go in more detail but I did not keep any detailed records on these conversations.
My take away is that it’s great at doing boilerplate not so great at uncommon tasks.
For example ask ChatGPT how to get the computer name in Powershell. The cmdlet it reccomends doesn’t exist.
When I ask ChatGPT to enumerate the displays connected to my graphics card it uses the correct cmdlet but then when I ask it to write a program to get the EDId data it makes up an API that doesn’t exist. When confronted it says you’re right that doesn’t exist.
In the case of JupyterLab it confuses Command Pallet with Command Registry and again says that there are functions called enable() and disable() which is incorrect.
The worse was when I was trying to ask about some boilerplate image processing stuff. I wanted to do a least squares fit and it completely made up APIs or recommended used libraries 10 years out of date.
I wish I could go in more detail but I did not keep any detailed records on these conversations.
My take away is that it’s great at doing boilerplate not so great at uncommon tasks.