1. It’s so so lazy - always weak answers that are not true and validated. 2. Always trying to take shortcuts - always encourages the “quick” way forward even though that is not the right direction for the project. 3. Constantly insists on taking breaks and closing the session - and if choosing to continue it becomes absent and quality of work is horrible. 4. Unbearable amount of mistakes - simple coding errors, core foundational strategies are thumb sucked and invalidated only to have an auditor flag and repeat the endless cycle. 5. Simply does not obey hooks in Claude.md 6. In general just a weak model that can’t think progressively and interpret and synthesis. 7. One million context window is the biggest joke - after 200k it starts to hallucinate and quality degrades exponentially. — before anyone says “prompt correctly” I’ve used Claude Mac app and Chat GPT app to create detailed prompts and hooks. I’ve truly tried to make it work.
@marcelzimmer@ClaudeDevs Yes!! Me too! And if it tell it that I set the pace and let’s continue - it becomes so lazy and dismissive. Results become weak. I’ve even tried compacting early with explicit compact summaries and it does not help.
@Identity_Crisiz@ClaudeDevs I swear I’ve tried everything. I’ve even used the Claude desktop app to write detailed prompts and Claude to create a detailed Claude.md file