2 ms·
Given this new information about the author's practices I'm inclined to think that everything they say in the above article should be treated with suspicion, es
by YeGoblynQueenne 6y ago
Given this new information about the author's practices I'm inclined to think that everything they say in the above article should be treated with suspicion, especially since they accept that a) they do not have ready access to GPT-3 and b) they edited the "GPT-3" article to some extent. For example I would not be surprised to learn that, unable to use GPT-3 for a significant amount of time, they genereated some of the text of the article and then created the rest by hand, or similar. Not because I don't believe that readers on the internet can mistake GPT-3 as something written by a human, but because the author's ethics sound just... dodgy.
- treis 6y agoMaybe they're running a deep con where they're fooling us into believing they fooled us
- YeGoblynQueenne 6y agoTruly, you have a dizzying intellect [1]. ___________ [1] https://youtu.be/rMz7JBRbmNo https://youtu.be/rMz7JBRbmNo
- minimaxir 6y ago> they edited the "GPT-3" article to some extent. For example I would not be surprised to learn that, unable to use GPT-3 for a significant amount of time, they genereated some of the text of the article and then created the rest by hand, or similar. You should assume all GPT-3 related text (including app demos and screenshots) is cherry-picked unless you can see the text generated in real time for yourself.
- YeGoblynQueenne 6y agoCherry picked, yes, but the author says they also edited the "GPT-3" article to some extent. The extent to which this was done is what I'm wondering about.
- acje 6y agoGiven the algorithm; generate text, cherry pick, edit. We could make a random number generator pass the turing test. It is just a mater of processing power and time.
- dvfjsdhgfv 6y agoThat's also my take. Anyo can say anything about GPT-3, and it's impossible to negate, because (1) most people don't have access, (2) the authors rarely provide complete input data, so even those who have access can't verify these claims. Really, this whole "OpenAI" is an oxymoron and I hope a truly open alternative will appear.