Wisdom Of The Week

• GPT2 is not uniquely dangerous, but is rather the latest step in an inevitable trend. It may or may not lower the cost to generate believable text, but powerful organizations can already produce massive amounts of much better content with humans and other algorithms already and GPT2 is not out of their reaches anyway.
• Fighting low quality babble, AI generated or not, is and will remain critical, but I think moving trust into demonstrably flawed “anti babble” algorithms is the wrong way to go. Someday, systems worthy of our trust will exist. Today is not that day.
• Instead of training people to rely on flawed algorithms, we should help them find the correct places to place their trust, first and foremost by training their own critical thinking skills. The human brain is currently still by far the most powerful truth generating machine we know.
• To catalyze that adaption of humans, I think these methods should be spread as wide as possible, so no one can hide behind a false sense of security. To further this (and put my money where my mouth is), I am planning to release 1.5B to the public.

- GPT2, Counting Consciousness and the Curious Hacker