The way I understand it (based on some introspection and reading the experiences of other autistic people), it's not a matter of ability to process information but rather the inability to not process information. We don't have the innate ability to recognize what's important and what isn't, which hinders our ability to recognize that two situations are the same and should be handled the same way. Asking "why?" is an attempt at understanding the pattern so that we can generalize in the same way as other non-autistics instead of memorizing every individual situation.
Absence of moderation is in itself a form of censorship.
Two more questions need answering before these findings can become actionable:
- How do these two groups compared to a third group that can use both? ChatGPT is pretty useless on its own when correctness is important, but it improves a lot when you combine it with ways to verify its output.
- How much time and effort would this new group need to accomplish the same task? One of ChatGPT's strengths is being able to communicate a piece of information in many different ways, and in whatever order you ask of it. It's then much faster to verify or through a legitimate source than it is to learn from those sources in the first place.
Where did all your pipes and wiring go? What insulates the building?
Like father, like son.
Generation ability seems to be about the same as any other model. The advantage of normalizing flows is that operations are invertible. That allows you to not just generate samples, but also calculate the probability of a sample.
It looks like Laurent Dinh (dude who originally came up with normalizing flows) is one of the authors of this work.
The long-forgotten technique? Normalizing flows. It feels like it was just last year that they were all the rage. It's insane how fast the field is moving.
Take from one group of poor people to benefit another group of poor people? That doesn't sound like a good thing to me.
Oh, I misunderstood. I thought you said he endorsed Mamdani.
I like having a separate connector for audio because it gets a lot of use and this lots of wear from the constant plugging and unplugging, and I'm often moving around with the headphones plugged in. I don't want to have to worry about breaking something from doing this.
Small USB connectors tend to be the first point of failure in most of my devices, and a broken USB port would render a phone completely unusable. I don't want to take that risk.
The training data anthropomorphizes the LLMs, so you'll get the best results by doing the same.