Ask HN: Any way to ban GPT-4 and others from harvesting my content and data?

Re:Ask HN: Any way to ban GPT-4 and others from harvesting m... by falcolas on Saturday August 13, 2022 @08:54PM (#32455256)

ML models learn very differently from humans; take very different pieces of information away from what they process.

ML is effectively parroting bits and pieces of what it's trained on back based solely on the input - this has quite clearly been demonstrated by Github's CoPilot software. This is different from a person who has internalized a lesson and writes code not only based on the input, but their own existing knowledge beyond the glyphs and their order that they've read.

Perhaps once we move beyond this parroting, we can revisit the ethics of free-for-all ML modeling. It's more plagiarism than free use in its current state.

[ Parent | on HN ]

Re:Ask HN: Any way to ban GPT-4 and others from harvesting m... by tiborsaas on Saturday August 13, 2022 @09:50PM (#32455601)

If you know something for a fact, then are you also just parroting back? These models rely on confidence so what else should they return if they get a prompt they have a 99.9999% probability to be a best match?

There's a difference between getting a predictable response for writing a fast inverse square root function and generating a custom solution that perfectly fits your data structure which matches the given context.

[ Parent | on HN ]