Claude 101

TL;DR

Enable Claude's memory feature for a 28% improvement in performance.
Keep Claude's memory short and focused to avoid context bloat.
Edit previous prompts to correct bad answers, rather than arguing.
Disable the 'downgrade' feature to ensure consistent model performance.
Empty the instructions box and use clear, single-goal prompts.
Use separate projects for different clients and disable account memory within projects.
Integrate Claude with Google Workspace applications.
Utilize GitHub links in prompts to leverage code repositories for specific tasks.
Request Claude to create dashboards from data.
Ask for a 'spread' of diverse ideas, including unlikely options.
Avoid using the thumbs up/down buttons to prevent chats from being used for training.
Define clear 'finish lines' for Claude's tasks and specify conditions for stopping.
Avoid leading questions; prompt Claude to argue against your assumptions.
Request specific file types from Claude, including documents and videos.
Use a 'one prompt' structure with clear Goal and Context sections.
Communicate context verbally or via screenshots instead of lengthy text.
Start new chats for long conversations to manage tokens and cache effectively.