I used your idea of “a” vs. “an” as the basis of a GPT-3 experiment to show that GPT-3 indeed probably does do lookahead. Details are at https://www.reddit.com/r/GPT3/comments/k0mvf3/experiment_that_shows_that_gpt3_can_probably_plan/
Thanks for sharing!
You’re welcome, and thank you for your post also :). I posted an updated version of my experiment, which (hopefully) improves the logic of my prior experiment, at https://www.reddit.com/r/MachineLearning/comments/k2n3yv/d_an_experiment_that_shows_that_gpt3_can_plan/.
I used your idea of “a” vs. “an” as the basis of a GPT-3 experiment to show that GPT-3 indeed probably does do lookahead. Details are at https://www.reddit.com/r/GPT3/comments/k0mvf3/experiment_that_shows_that_gpt3_can_probably_plan/
Thanks for sharing!
You’re welcome, and thank you for your post also :). I posted an updated version of my experiment, which (hopefully) improves the logic of my prior experiment, at https://www.reddit.com/r/MachineLearning/comments/k2n3yv/d_an_experiment_that_shows_that_gpt3_can_plan/.