I wonder how model performance would change if the tooling included the ability to interact with the binary and validate the backdoor. Particularly for models that had a high rate of false positives, would they test their hypothesis?
At the end of the response they forget everything. They need to be fed the entire text for them to know anything about it the next time. That is not surpassing even feline intelligence.
This is backwards. When covid started, the executives were saying that working from home will not affect productivity. They said zoom and slack are the new way to work. Some even said that productivity increased when people didn't have to commute. Some said people were working longer hours than earlier.
Now, all of a sudden, they realize their blunder. While the workers are happy, they are not. So now they will try to convince us that we are not productive enough, or that some of us are slacking, or that we have stopped innovating, or another 50 reasons for which going to the office will help us.
More recently, they have realized that they are not fooling anyone. So now they are threatening lower pay or layoffs for those who refuse to come back.
I have tried a similar idea before. I had two experts: one to propose a plan and another to critique it. It worked well, but often required 5 rounds of iteration to get to a great answer. The cost wasn't ideal and GPT4 still performed better on the first go, so I abandoned the idea.
I am curious how this approach compares to one-shot response from a larger model.