> A larger question: Are we hurling ourselves toward a (near) future of unaligned AGI with self-improvement capabilities? We are running towards a brick wall a

> A larger question: Are we hurling ourselves toward a (near) future of unaligned AGI with self-improvement capabilities? We are running towards a brick wall and people are not paying attention. Setting up self-reflection loops today is actually fairly trivial and can be done programmatically, all the model needs is to produce a solution, invoke the evaluation and keep iterating.