Social Media & AI

Hi All,

Just a quick question for those interested in AI technology, specifically with Social Media.

With AI, what are the smartest methods for finding language that could be seen as offensive or not considerate of other cultures?

Also, what tools would suggest I use within Machine Learning?

I'm open to any suggestions or pointers.

info[at]senost[dot]com

That is an incredibly hard question, one they repeatedly failed to solve.

The latest is the large language models and even they are hard pressed and cannot catch all.

In the end a big part of the problem is that it’s subjective.

Sure, there is a lot that is obvious and that it would catch, but a lot is subtle and contextual and very very hard to distinguish from the good way and the same sentence can actually be both depending on context and the models are still only statistical models, they do not understand the words.

And then you have the manipulation problem where people bypass or trick the filter.

Mu guess is that a combination of different models together with some independent verifier “is this judgement correct” might overcome many of the problems.

But both the building of the models snd running multiple models will require lots of performance.

Creating dedicated LLM’s for this purpose could be the best but would cost millions or more to create and would require massive amount if manual work classifying examples to get good coverage.