So I just read this article

@oliviaintech

2.6K followers · usually 1.1K views a post

Views

4.6K

Likes

159

Comments

8

Shares

22

Selling · SaaS › AI Tool

Claudeseen in 999 videos →

Why it worked

The video provides timely and practical advice on optimizing interactions with a popular AI model, addressing a common pain point for users and offering actionable insights.

Summary

The video discusses new methods for interacting with Claude AI models, based on an article by one of its engineers. It suggests simplifying prompts and using auto-memory instead of explicit rules or large memory files to improve Claude's performance.

Structure

1Introduction to the article and its author
2Explanation of previous methods for correcting Claude
3Introduction of new methods for working with Claude models
4Specific advice on prompt simplification and auto-memory
5Call to action for viewer engagement

Product placement

Claude: an AI model that generates text and code. It ACTS as the subject of the video. Removing it would change the video's topic. Thariq: a person who built Claude Code. He APPEARS as the author of an article. Removing him would not change the video's content.

Call to action

Original caption

With the Claude 5 models there is a new way to be working with Claude #claude #vibecoding #chatgpt

More from @oliviaintech

Their other posts in the index, biggest breakout first.

let me know what you think of them in the comments.

On-screen text

Give-Claude Rules Give-Claude Examples Put-it-all-upfront Repeat Yourself Memory-in-Claude.MDs Simple-Specs → Give Claude Judgement → Design Interfaces → Use Progressive Disclosure → Simple Tool Descriptions → Auto-memory → Rich References

Transcript

So I just read this article by Thariq, one of the engineers building Claude Code. And apparently, with the release of the new Claude models, there is a completely new way we should now be working with Claude. Previously, when Claude would get something wrong, you'd probably try putting in something in the system prompt or the MD file, or maybe you try giving it an explicit example. But with these new models, apparently more information and especially contradicting rules can actually constrain how much Claude explores. And it can even get confused and spend a lot of extra time thinking. So, apparently, the new way of working with Claude is to simplify and provide only the information that Claude actually needs and provide it when Claude needs it. So try to avoid explicit rules. Try to avoid repeating yourself and instead stuffing a memory MD file, use auto memory instead. I'm curious to actually start trying these things, but let me know what you think of them in the comments.