one thing coming up in the debate about the pause letter i really agree with: openai should make a great alignment dataset and alignment evals and release those! bonus points if we can find a prototype democratic process for 'what we align to'.
@sama 100% alignment is a mirage (and 100% is required). Look at the U.S. tax code. There are always loopholes.
@sama Use Chainlink to validate and ensure the process can’t be tampered with
@sama What do you think about it? @DogmanLabs @VampishBeef @grahappa @DanSingjoy @velua @wildwex @joshuaseymour @gofractally @_JamesMart @prayfractally @amiHeines @mschoenebeck1 @bytemaster7
@sama An artificial pause for 6 months wouldn’t accomplish a thing. There should be a way for debating publicly about the ethics of AI/AGI instead of having small committee of luminaries decide for everybody else.
I think we misunderstand something about democracy... One of the main benefits of democracy is that diverse groups can compromise on a path forward. But that doesn't mean that you'd normally want those groups to lose their identities or agree to a single common set of values. One of the benefits of that individuality is that as circumstances change and the optimal path forward changes, some groups can gain influence while others lose it. A minimum common standard would be more akin to a set of laws. But laws (and their enforcement systems) cover very little of human behavior. That laws are not always fully followed is also not always a bad thing - government doesn't always know what is best for the people, and if people adapt around problems that government actions generate - that can make the system healthier as a whole. ... I think we've come to the point where we need to start understanding things that humanity has in fact, never understood clearly. And with tools that can simulate human like conduct, like GPT-4, that may even be possible.
@sama I wonder how many people have been promoting this kind of solution.
@sama Ser gpt 4 making basic math wrong (addition and multiplication)