Conversation

Notices

1. Embed this notice
   藤井太洋, Taiyo Fujii (taiyo@ostatus.taiyolab.com)'s status on Monday, 03-Apr-2023 08:55:13 JST 藤井太洋, Taiyo Fujii

   GPTは盗んだ著作物を発表することもないし、労働者を解雇することも、デマを発信することもない。公開書簡で懸念されたことをやるのは人間だ。
   規制は必要だ。学習した対象を公開する必要はあるし、それぞれの出力に関与した学習元を提示できるように要請してもいい（おそらく関与元が手繰れるようになるとカスタマイズがものすごく容易になる）。開発会社が倫理委員会を結成し、事象を再現可できるサンドボックスを作るのもいい。

   しかし停止はナンセンスだ。仮にOpenAIが停止要請に応じたとして、モラトリアムが開けた後に私たちが見るものは、停止に応じない「悪いAI」と、停止を要請した人の何人かが出資したAIがGTP-5に肩を並べて集金効率を競い合う姿。

   • Embed this notice
     藤井太洋, Taiyo Fujii (taiyo@ostatus.taiyolab.com)'s status on Monday, 03-Apr-2023 09:15:33 JST 藤井太洋, Taiyo Fujii

     in reply to

     AIを利用してデマをばらまいたり、著作権を侵害したりする人や企業をリストアップする活動を始めるとかね。出力物を照合できるAIサービス企業にしかできないこともある。

   • Embed this notice
     藤井太洋, Taiyo Fujii (taiyo@ostatus.taiyolab.com)'s status on Friday, 14-Apr-2023 18:42:46 JST 藤井太洋, Taiyo Fujii

     in reply to

     • Dekkzz :emacs:

     @dekkzz76 GTP produce stolen article, but they don’t publish. We, human do publish.

   • Embed this notice
     Dekkzz :emacs: (dekkzz76@emacs.ch)'s status on Friday, 14-Apr-2023 18:42:47 JST Dekkzz :emacs:

     in reply to

     @taiyo

     your first para is correct bar 1 item, GPT does publish "stolen" works, it's how it was trained.

   • Embed this notice
     藤井太洋, Taiyo Fujii (taiyo@ostatus.taiyolab.com)'s status on Friday, 14-Apr-2023 19:22:36 JST 藤井太洋, Taiyo Fujii

     in reply to

     • Dekkzz :emacs:

     @dekkzz76 I understand what you point. Somebody publish and let piracy article flood. Results would be same.

     But for me, I’m a writer, and the publisher. I am aware of differences of those. The motivation of those are close but differ. And more, responsibility is not the same. The publisher takes all the responsibility.

     And I hope GPT and Bart to be capable showing all the references articles of each output with effective rate. And re-producing with choosing sources.

   • Embed this notice
     Dekkzz :emacs: (dekkzz76@emacs.ch)'s status on Friday, 14-Apr-2023 19:22:37 JST Dekkzz :emacs:

     in reply to

     @taiyo

     But the effect is the same, a stolen article gets published, then like word of mouth is picked up by other GTP AI & is then given legitimicy.

     In Bing or Google, an alogrithm publishs GTP output after a request for information, does that blur the lines? Is there a "human" element to the display or creation of that output?

   • Embed this notice
     藤井太洋, Taiyo Fujii (taiyo@ostatus.taiyolab.com)'s status on Friday, 14-Apr-2023 19:24:03 JST 藤井太洋, Taiyo Fujii

     in reply to

     • Dekkzz :emacs:

     @dekkzz76 Controlling those with source and disclosing references let LLMs to be the real-tool of us.

   • Embed this notice
     藤井太洋, Taiyo Fujii (taiyo@ostatus.taiyolab.com)'s status on Friday, 14-Apr-2023 20:20:11 JST 藤井太洋, Taiyo Fujii

     in reply to

     • Dekkzz :emacs:

     @dekkzz76 I already write with GPT for English, not pro writing. And not long sentence but part by part, words by words. And I rewrite GPT output further transformative as I believe my own writing.

     I’m trying to record how many GPT output is left in my draft. On my latest writing, 86 % was changed from GPT by editing distance.

   • Embed this notice
     Dekkzz :emacs: (dekkzz76@emacs.ch)'s status on Friday, 14-Apr-2023 20:20:12 JST Dekkzz :emacs:

     in reply to

     @taiyo

     so will GPT do your writing for you?