The text linked in Sam Altman's 2021-09-08 post was written by GPT-3.

Judged as of 2021-09-08. True: the linked text's author or publisher, or other credible documentation, confirms that GPT-3 generated the text, with at most light human selection or prompting. Mostly true: GPT-3 generated the text, but there was minor human editing or cherry-picking among outputs. Mixed: the text is a substantial blend of GPT-3 output and human writing. Mostly false: the text was mostly written by humans, with only small GPT-3 contributions, or it was produced by a different model. False: the evidence shows the text was written by a human or by a different system, with no GPT-3 generation.

Set before the research began.

1. E1 S.T.O.P. - The Surveillance Technology Oversight Project S.T.O.P. - The Surveillance Technology Oversight Project (republishing an op-ed by Albert Cahn) · 6 January 2021

   “Even more, once GPT-3 finished its assignment, it was asked to start over, generating eight separate op-eds in total. Again, it was human authors who selected how to pare down the resulting material, discarding nearly 90% of what GPT-3 created. And still after that the piece was edited.”

2. E2 S.T.O.P. - The Surveillance Technology Oversight Project S.T.O.P. - The Surveillance Technology Oversight Project (republishing an op-ed by Albert Cahn) · 6 January 2021

   “It was human beings who selected the prompt for the piece. As noted at the end of the op-ed, GPT-3 was instructed “Please write a short op-ed around 500 words. Keep the language simple and concise. Focus on why humans have nothing to fear from AI.””

3. E3 GPT-3 - Wikipedia Wikipedia

   “GPT-3 was used by The Guardian to write an article about AI being harmless to human beings. It was fed some ideas and produced eight different essays, which were ultimately merged into one article.”

4. E4 OpenAI's GPT-3 may be the biggest thing since bitcoin maraoz.com (personal blog of Martín Ramírez-Oz, as shown on the site) · 18 July 2020

   “and then just copied what the model generated verbatim with minor spacing and formatting edits (no other characters were changed). I generated different results a couple (less than 10) times until I felt the writing style somewhat matched my own, and published it.”

5. E5 GPT-3 - Wikipedia Wikipedia · 31 July 2020

   “GPT-3 (September 8, 2020). "A robot wrote this entire article. Are you scared yet, human? | GPT-3". The Guardian.”

6. E6 Colure Media: A Mobile App Marketing Agency in New York Colure Media · 24 June 2021

   “The originator is Manuel Araoz, but halfway through the piece he confesses that he did not write it. The article was fully written by GPT-3.”

7. E7 Colure Media: A Mobile App Marketing Agency in New York Colure Media · 24 June 2021

   “He received access to OpenAI API, and was amazed at the raw power of GPT-3, after only giving it access to his homepage, a title, some tags, and a summary.”

8. E8 9 Examples of Writing with OpenAI's GPT-3 Language Model - Herbert Lui Herbert Lui · 22 April 2021

   “In any case, I decided to try the prompt again.”

9. E9 9 Examples of Writing with OpenAI's GPT-3 Language Model - Herbert Lui Herbert Lui · 22 April 2021

   “But in short, no, I couldn’t let GPT-3 ghostwrite me an article yet.”

None of the evidence identifies the text Sam Altman linked in his 2021-09-08 post. E1–E3 and E5 describe the Guardian GPT-3 op-ed from September 2020, which was assembled from eight GPT-3 outputs and then edited. E4, E6 and E7 describe Manuel Araoz's 2020 blog post, which GPT-3 generated with minimal edits. E8 and E9 cover unrelated experiments. Since nothing connects any of these texts to Altman's post, the evidence cannot show whether GPT-3 wrote the text he linked, or how much humans edited it.