Do we want to train the model during inference? I would think we want to be very particular about the data used to train - notably after OpenAI's notes on the subject.
What happens when you hit a guardrail because there was a politically-charged or socially-condoned phrase? What if the documents are potentially sensationalist news articles or Wikipedia articles with potentially-questionable topics?
“Disclaimer: Richard Feynman was a colossal misogynist. I have a hard time separating this fact from his contributions to science.”
I have heard that Socrates was fond of the young boys whom he taught. Why mention an irrelevant bias of one and not another? Is the point to condone behaviour that isn’t yet fully condoned by present society?
How do all the commenters here know that the images are eliciting a fear response? The referenced study measures firing rates of visual neurons. It does not discuss ultimate emotional reactions.
- Dove head-first into the fundamentals of distributed systems, applying all that I learned to solutions for scalability issues at work.
- Realized that the secret to promotion is that it’s all about making your superiors look good. This brought on even greater realizations about society in general and individuals in my life.
- The importance of work/life balance and that the “life” portion must involve taking good care of yourself.
- The value of being kinder to all those I see regularly. It’s easy to drop some of the positive interactions with those we’re comfortable with.
Can we measure when he loses count by using many small scoops? I expect if his current bowl has passed a certain threshold of scoops he won't bother switching no matter how many scoops are going there.