I was surprised to find that you.com makes the same mistakes, e.g. treating “SolidGoldMagikarp” as “distribute”. I wouldn’t have expected two unrelated systems to share this obscure aspect of their vocab...
Why might this be? It could be ChatGPT and you.com chat share a vocabulary. It could be they use a similar method for determining their vocab, and a similar corpus, and so they both end up with SolidGoldMagikarp as a token. It could be you.com’s chat is based on ChatGPT in some other way. Maybe you.com is using GPT-J’s public vocab. I would be interested to know if the vocab overlap is total or partial.
I was surprised to find that you.com makes the same mistakes, e.g. treating “SolidGoldMagikarp” as “distribute”. I wouldn’t have expected two unrelated systems to share this obscure aspect of their vocab...
Why might this be? It could be ChatGPT and you.com chat share a vocabulary. It could be they use a similar method for determining their vocab, and a similar corpus, and so they both end up with SolidGoldMagikarp as a token. It could be you.com’s chat is based on ChatGPT in some other way. Maybe you.com is using GPT-J’s public vocab. I would be interested to know if the vocab overlap is total or partial.