General Discussion

Personally, in 1995, I created an experimental little app for Windows. Called FLAMER. It was a satirical AI engine. What it did was read in messages from discussion sites

Then. It looked for keywords in a specific database It held in memory.. That database was large, but fixed. I populated the data, or it could be loaded from a file.

Next, it wrote a reply to the message it had read in. The reply specifically insulted the writer of the message. That used a language module that selected the words, phrases, and structures stored in a language database, also loaded from a file. A small language model, but large enough. That, too, was flexible and new files could be created.

For speed, all data was stored in memory in matrices.

Finally, a randomized insulting reply was generated from the SLM, and could be pasted into the thread by the user. Internal checking prevented duplications and so on.

Did it work? Pretty well. Well enough that the temporary users I used to experiment were often banned from the forums I used for testing. It was a joke program. A venture into AI.
I got bored, though, so I released the coding into the public domain.