I'd have to disagree with Gary. There has been significant improvement in the ability to attend to tokens in context, resulting in fewer missed instructions and facts. For instance, Claude 3.5 Sonnet appears to capture every detail, whereas GPT-4 often overlooks numerous

→ View original post on X — @marek_rosa