Peter Norvig critically reviews AlphaCode's code quality

Re:Peter Norvig critically reviews AlphaCode's code quality by sdenton4 on Friday December 16, 2022 @06:38PM (#34022188)

Sub-token manipulation is a known weakness of the models; they are fed blocks of characters, not individual characters, and so they underperform on tasks that require fine-character-level manipulations.

[ Parent | on HN ]

Re:Peter Norvig critically reviews AlphaCode's code quality by kqr on Saturday December 17, 2022 @04:58AM (#34026438)

But in that case I would have wanted to see it present a number that's at least of the right order of magnitude, even if all individual digits were incorrect. A human can adapt to their known weaknesses.

[ Parent | on HN ]

Re:Peter Norvig critically reviews AlphaCode's code quality by sdenton4 on Saturday December 17, 2022 @11:59AM (#34029610)

Optical illusions expose fundamental issues in human perception, which we generally can't adapt to. This is similar.

[ Parent | on HN ]