Skip to main content
Role-playing Games

Return to Answer

typo fix
Source Link
Valley Lad
Valley Lad
  • 7.7k
  • 7
  • 9

As a 20+ year AI developer, I say, Yes, we should ban it.

I have never touted my background on an any answer on this stack before, but I want to say, I have done not just AI generally but natural language processing in particular, for a lot of companies, and I have over 20 issued (not just pending) US patents in this area -- and I say we need to ban it, because of how it works.

ChatGPT uses RLHI (Reinforcement Learning with Human Inputs). A number of systems like this are in the works from several companies. Each one is a statistical "big data" approach, through which it harvests snippets (phrases) from sentences all over the web (and/or from directly collected human responses), then parses them (first syntactically, then semantically), and finally, reassembles them to make an answer -- then waits to see if the user who asked it the question gives good or bad feedback.

The positive impact of that feedback is slow and gradual, at best, and depends on the quality and reliability of the feedback. The system tends to take no feedback as a soft confirmation that it is not wrong.

Until and unless a very strong domain-specific model is built within ChatGPT that keys directly into RPG systems (it would literally have to train, for example, on this stack) -- it will not be reliable. And if it did train on this stack, we would notice it - we'd see phrases from past answers getting parroted in bits and pieces in the new answers it creates.

And BTW, similar systems are coming from Microsoft and Google, and possibly later from Facebook and Apple, so, when we write the policy, it shouldn't say "ChatGPT", it should say something like "machine learning-based chat bots including but not limited to ChatGPT".

As a 20+ year AI developer, I say, Yes, we should ban it.

I have never touted my background on an any answer on this stack before, but I want to say, I have done not just AI generally but natural language processing in particular, for a lot of companies, and I have over 20 issued (not just pending) US patents in this area -- and I say we need to ban it, because of how it works.

ChatGPT uses RLHI (Reinforcement Learning with Human Inputs). A number of systems like this are in the works from several companies. Each one is a statistical "big data" approach, through which it harvests snippets (phrases) from sentences all over the web (and/or from directly collected human responses), then parses them (first syntactically, then semantically), and finally, reassembles them to make an answer -- then waits to see if the user who asked it the question gives good or bad feedback.

The positive impact of that feedback is slow and gradual, at best, and depends on the quality and reliability of the feedback. The system tends to take no feedback as a soft confirmation that it is not wrong.

Until and unless a very strong domain-specific model is built within ChatGPT that keys directly into RPG systems (it would literally have to train, for example, on this stack) -- it will not be reliable. And if it did train on this stack, we would notice it - we'd see phrases from past answers getting parroted in bits and pieces in the new answers it creates.

And BTW, similar systems are coming from Microsoft and Google, and possibly later from Facebook and Apple, so, when we write the policy, it shouldn't say "ChatGPT", it should say something like "machine learning-based chat bots including but not limited to ChatGPT".

As a 20+ year AI developer, I say, Yes, we should ban it.

I have never touted my background on any answer on this stack before, but I want to say, I have done not just AI generally but natural language processing in particular, for a lot of companies, and I have over 20 issued (not just pending) US patents in this area -- and I say we need to ban it, because of how it works.

ChatGPT uses RLHI (Reinforcement Learning with Human Inputs). A number of systems like this are in the works from several companies. Each one is a statistical "big data" approach, through which it harvests snippets (phrases) from sentences all over the web (and/or from directly collected human responses), then parses them (first syntactically, then semantically), and finally, reassembles them to make an answer -- then waits to see if the user who asked it the question gives good or bad feedback.

The positive impact of that feedback is slow and gradual, at best, and depends on the quality and reliability of the feedback. The system tends to take no feedback as a soft confirmation that it is not wrong.

Until and unless a very strong domain-specific model is built within ChatGPT that keys directly into RPG systems (it would literally have to train, for example, on this stack) -- it will not be reliable. And if it did train on this stack, we would notice it - we'd see phrases from past answers getting parroted in bits and pieces in the new answers it creates.

And BTW, similar systems are coming from Microsoft and Google, and possibly later from Facebook and Apple, so, when we write the policy, it shouldn't say "ChatGPT", it should say something like "machine learning-based chat bots including but not limited to ChatGPT".

Source Link
Valley Lad
Valley Lad
  • 7.7k
  • 7
  • 9

As a 20+ year AI developer, I say, Yes, we should ban it.

I have never touted my background on an any answer on this stack before, but I want to say, I have done not just AI generally but natural language processing in particular, for a lot of companies, and I have over 20 issued (not just pending) US patents in this area -- and I say we need to ban it, because of how it works.

ChatGPT uses RLHI (Reinforcement Learning with Human Inputs). A number of systems like this are in the works from several companies. Each one is a statistical "big data" approach, through which it harvests snippets (phrases) from sentences all over the web (and/or from directly collected human responses), then parses them (first syntactically, then semantically), and finally, reassembles them to make an answer -- then waits to see if the user who asked it the question gives good or bad feedback.

The positive impact of that feedback is slow and gradual, at best, and depends on the quality and reliability of the feedback. The system tends to take no feedback as a soft confirmation that it is not wrong.

Until and unless a very strong domain-specific model is built within ChatGPT that keys directly into RPG systems (it would literally have to train, for example, on this stack) -- it will not be reliable. And if it did train on this stack, we would notice it - we'd see phrases from past answers getting parroted in bits and pieces in the new answers it creates.

And BTW, similar systems are coming from Microsoft and Google, and possibly later from Facebook and Apple, so, when we write the policy, it shouldn't say "ChatGPT", it should say something like "machine learning-based chat bots including but not limited to ChatGPT".