“After deploying GPT-3, OpenAI found that the creation of "spammy content" was a much larger risk than the anticipated risk of election misinformation.”