Which is logical and kind of what I expected. But raises the obvious question of where does your data come from going forward? The internet is getting more and more polluted with machine generated data, previous big ongoing data sources like Twitter, Reddit, etc. are all full of GPT spam and are trying to monetise their data.

I’d also be interested in how much you think your platform is just capturing say reported surveys/data. President polling is something that must be all over LLM datasets- isn’t that just replicating the training data?

I think you could do a better job of showing on your website the following - here are some unusual survey results we generated from the model - I.e. stuff definitely not in the training data - and here’s the data we actually got when we did that survey for real