What intrigues me is this....If the Founding Fathers and so on are portrayed as black people, including black females, and these images becomes credible to newer generations,
How will the entire borg of white people as enslavers and colonial settlers be maintained? I know this sounds far fetched, but it looks like this thing is on a huge collision with its own ideologies.
The good news is that this seems easy enough to fix, just optimize the model for accuracy. The classic example is GPT-4’s paper that showed GPT-generated probabilities vs observed probabilities before and after RLHF, with massive degradation of accuracy induced.
So, who will build I model that isn’t trained to be wrong? Is it reasonable to have hope for the new release of Grok?
Great breakdown of how Google botched their release. I say botched but it was deliberate, with unintended consequences once people noticed what was going on.
I've come to realize that AI Misalignment won't be the problem with AI. Human Misalignment with other humans is the problem. This is nothing new, misalignment the human condition, one group of humans is misaligned with any/every other group of humans and are going to use the available tools to leverage their misalignment.
The Ukraine War is a human to human misalignment. The UK Horizon Post Office Scandal was a human to human misalignment—really the UK Post Office scandal was several human to human misalignments.
What is really interesting here is that Google is misaligned with the truth of the past. Google is The Ministry of Truth.
What intrigues me is this....If the Founding Fathers and so on are portrayed as black people, including black females, and these images becomes credible to newer generations,
How will the entire borg of white people as enslavers and colonial settlers be maintained? I know this sounds far fetched, but it looks like this thing is on a huge collision with its own ideologies.
The good news is that this seems easy enough to fix, just optimize the model for accuracy. The classic example is GPT-4’s paper that showed GPT-generated probabilities vs observed probabilities before and after RLHF, with massive degradation of accuracy induced.
So, who will build I model that isn’t trained to be wrong? Is it reasonable to have hope for the new release of Grok?
Great breakdown of how Google botched their release. I say botched but it was deliberate, with unintended consequences once people noticed what was going on.
I've come to realize that AI Misalignment won't be the problem with AI. Human Misalignment with other humans is the problem. This is nothing new, misalignment the human condition, one group of humans is misaligned with any/every other group of humans and are going to use the available tools to leverage their misalignment.
The Ukraine War is a human to human misalignment. The UK Horizon Post Office Scandal was a human to human misalignment—really the UK Post Office scandal was several human to human misalignments.
What is really interesting here is that Google is misaligned with the truth of the past. Google is The Ministry of Truth.