4 ms·

Do you consider it theft because of the scale? If I read something you wrote and use most of a phrase you coined or an idea for the basis of a plotline in a boo

Do you consider it theft because of the scale? If I read something you wrote and use most of a phrase you coined or an idea for the basis of a plotline in a book I write, as many authors do, currently it's counted as being all my own work. I feel like the argument is akin to some countries considering rubbish, the things you throw away, to still be owned by your person ie "dumpster diving" is theft. If a company had scraped public posts on the Internet and used it to compile art by colourising chunks of the text, is it theft? If an individual does it, is it theft?

ToucanLoucan 2y ago
This argument has been stated and re-stated multiple times, this notion that use of information should always be free, but it fails to account for the fact that OpenAI is not consuming this written resource as a source of information but rather as a tool for training LLMs, which it has been open about from the beginning is a thing it wishes to sell access to as a subscription service. These are fundamentally not the same. ChatGPT/Copilot do not understand Python, they are not minds that read a bunch of python books and learned python skills they can now utilize: they are language models, that internalized metric tons of weighted averages of python code and can now (kind of) write their own, based on minimizing "error" relative to the code samples they ingest. Because of this, Copilot has never and will never write code it hasn't seen before, and by extension of that, it must see a whole lot of code in order to function as well as it does. If you as a developer look at how one would declare a function in python, review a few examples, you now know how to do that. Copilot can't say the same. It needs to see dozens, hundreds, perhaps thousands of them to reasonably accurately be counted on to accomplish that task, it's just how the tech works. Ergo, scaled data sets that can accomplish this teaching task now have value, if the people doing that training are working for high-valuation startups with the objective of selling access to code generating robots.

ToucanLoucan 2y ago

This argument has been stated and re-stated multiple times, this notion that use of information should always be free, but it fails to account for the fact that OpenAI is not consuming this written resource as a source of information but rather as a tool for training LLMs, which it has been open about from the beginning is a thing it wishes to sell access to as a subscription service. These are fundamentally not the same. ChatGPT/Copilot do not understand Python, they are not minds that read a bunch of python books and learned python skills they can now utilize: they are language models, that internalized metric tons of weighted averages of python code and can now (kind of) write their own, based on minimizing "error" relative to the code samples they ingest. Because of this, Copilot has never and will never write code it hasn't seen before, and by extension of that, it must see a whole lot of code in order to function as well as it does. If you as a developer look at how one would declare a function in python, review a few examples, you now know how to do that. Copilot can't say the same. It needs to see dozens, hundreds, perhaps thousands of them to reasonably accurately be counted on to accomplish that task, it's just how the tech works. Ergo, scaled data sets that can accomplish this teaching task now have value, if the people doing that training are working for high-valuation startups with the objective of selling access to code generating robots.