5). Scaled-up Instructable Model Become Less Reliable – suggests that larger and more instructable LLMs may become less reliable; investigates LLMs across three elements: difficulty concordance, task avoidance, and prompting stability…

→ View original post on X — @dair_ai