Closing thoughts is where I give my (somewhat divergent, "bigger picture") conclusion to this post. Seed prompt is to be run as part of a prompt with qualifications, an example (qualifying input, output) pair [1] is (standardization-example,Standardization-response-1.pdf). Analysis, a more important section comments on the latter document. Some highlights are...
The point. We all are confidently overlooking the root cause of why the alignment problem exists and I claim to have found a cluster of reasons but I cannot convince the right (amount) of people to take them seriously to make any progress on it. It relates to ML-involving technology...
Epistemic status. Every idea that this post builds on moved into the center of my focus while studying and working on projects related to AI alignment, especially in last year's AIS Camp. Afterwards, as I drifted further from practical or strategic AIS studies, I was trying to put them into...