VERIFIABILITY IS THE LIMIT

Summary from the TLDR newsletter

There are many beliefs about the limits, usefulness, and potential of large language models. Every piece of software needs to be 'correct', which means that the software conforms to the intent of the creator. The creator has to verify such conformance - it cannot be offloaded. Verifiability is the limit that tells us what we can't implement via large language models, and it cannot be solved with agentic approaches.