Excuse me, a typo in my initial tweet. I meant: a live action shot of SHRDLU, from 1970: https://en.wikipedia.org/wiki/SHRDLU
GPT-3 is wonderful, & I'm fascinated by all the examples. But be careful in how you reason about what is new.
GPT-3 is wonderful, & I'm fascinated by all the examples. But be careful in how you reason about what is new.
Fun fact, the developer of SHRDLU was Terry Winograd, at the time a grad student. Several decades later he was a professor at Stanford, supervising a grad student named Larry Page...
This thread is sorta a weird reverse Turing test: can you distinguish your modern AI system from a 1970-era AI system? If the answer is "no", does that mean your modern AI is no smarter than the 1970 system?
(The answer, of course, is clearly "no". But the reasons are at least somewhat complicated...)
And @joshu uses GPT-3 to imagine a 3-way conversation between a person, SHRDLU, and GPT-3
https://twitter.com/joshu/status/1283995509619740673

Now SHRDLU is getting introspective: https://twitter.com/flantz/status/1284001806666928136
I feel like I _almost_ recognize the writing style. Hmm.
I feel like I _almost_ recognize the writing style. Hmm.
Kudos for the honesty: https://twitter.com/TheFrontalLobe_/status/1284003086831366146
I enjoyed this, on GPT-3 alone: https://twitter.com/zebulgar/status/1283927560435326976