rlprlprlp

This frustrates me as well. Especially when I set a goal and come back, ask what went wrong, and the model hits me with something like “sorry i didn’t do that, gave up when x happened” and then I ask if x actually happened and the response is like “oops no I just assumed.” I had similar issues with older versions of sol at the ultra effort level. Lowering it seemed to help then, but I wish I had better insight for getting Astra to work better without e.g. having another LLM check its progress every hour.

paths friend. so first turn you have ten options, second u have the same ten. assume turn one model h1. turn two you have ten options. 1x10. but you could have chosen any h on the first turn. so 10x10=100. turn three you could have gotten there 100 ways and still have ten options.

i have no take on this comment, but i like the model in this paper for considering whether it could be possible or appropriate wrt different levels of vetting:

https://arxiv.org/pdf/1801.03533

Sure, but you’re god deciding who goes to hell.