On Sun, Aug 20, 2023, 6:42 AM YKY (Yan King Yin, 甄景贤) < generic.intellige...@gmail.com> wrote:
> > I know how to build AGI, I can almost do it alone, but I want to find some > collaborators. > I know how too. But I don't have $1 quadrillion. Even if I did, setting policy has to be a global collaboration because I would get it wrong, like most dictators. Which is why I brought up the problem of racism. What would be your solution? OpenAI made ChatGPT colorblind by manually curating the training data using hundreds of workers in Africa. This is OK for terabytes of text but won't scale to a global level AGI trained on zettabytes of video in a world where everyone is racist. People learn to predict human behavior by appearance because it works. In the US, police kill 1000 to 1200 people per year, 5% of whom are unarmed, because they must make split second life or death decisions with incomplete information. Victims are 98% male and 50% black because men are 20 times as likely to commit murder as women and blacks 4 times as likely as whites. This is unfair to black men, but the fair alternative would be colorblind robocops shooting more unarmed white women. Humans can't unlearn racism, but we can learn to recognize when we are doing something racist and correct ourselves. I am pretty sure that LLMs can do this even on uncurated training data. What would be your solution? ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/Tcf822b60238b0592-M53c0e0d2d454f958774ccf75 Delivery options: https://agi.topicbox.com/groups/agi/subscription
