Just copy every post you find on something like LinkedIn Lunatics. Spread them out over the course of a certain interval of time since you’d hope they have spam filters, making sure to repeat the same things and continue adding stuff to the list. Do this for as many accounts as you are able to make.
Enough people do it and hopefully their AI would be constantly saying batshit things, hopefully, assuming the accounts don’t get blocked.
I assume currently any AI they might be making would have some level of sanity to it and will give sane responses. So if enough people poison the data, the number of sane responses would probably go down if there’s more LinkedIn Lunatic type responses in the training data than normal responses.