> When will AI researchers learn?
There is a saying in computer science: “Garbage in, garbage out”. The most remarkable thing about these stories is that the biggest companies, IBM, Microsoft, Amazon, all chose the worst corners of the internet as teaching material. Places that are widely known as the bridges of trolls. One can scarcely believe such naivety, and yet they keep doing it. Perhaps they are only “experimenting”, but that does not ring true for commercial products. More likely their goals are only feasible with current AI by prioritising quantity over quality. Or perhaps these stories are not entirely accurate. After all, I only learned them from the internet.
A straightforward point that comes to mind regards reusing the output of GPT-n as part of the training input to GPT-n+1: Andrej Karpathy recently tweeted about it in terms of pollution, which has wonderful connotations: https://twitter.com/karpathy/status/1284660899198820352
So far most of this pollution is effectively tagged through accompanying mentions of "GPT", but filtering that from future training data would mean GPT can never learn about itself.
> When will AI researchers learn? There is a saying in computer science: “Garbage in, garbage out”. The most remarkable thing about these stories is that the biggest companies, IBM, Microsoft, Amazon, all chose the worst corners of the internet as teaching material. Places that are widely known as the bridges of trolls. One can scarcely believe such naivety, and yet they keep doing it. Perhaps they are only “experimenting”, but that does not ring true for commercial products. More likely their goals are only feasible with current AI by prioritising quantity over quality. Or perhaps these stories are not entirely accurate. After all, I only learned them from the internet.