The Evaluation Loop: How AI Improves Its Own Output

AI is better at evaluating content than generating content.​
​
This makes sense, when you stop to think about it. For a large language model to interpret your instructions, and create output, all in the same step, it dilutes the same token budget over two requests.
​
But if you have ask AI to evaluate its own outputs, it can focus all its attention on identifying how to improve. Then it can rewrite, separately.
​
Here’s a simple copy/paste process you can use to rapidly improve your LLM responses:

1. Ask an LLM to create some content for you
2. In a second tab, have another LLM evaluate the output.
3. Ask the second LLM for a clear brief documenting improvements.
4. Copy that brief, and paste it into the first LLM tab with the prefix, “Analyse this feedback.”
5. Ask for a V2.
6. Repeat until satisfied
7. Compare with diffchecker.com if you want to see the changes

This simple process will help you stir your prompt around in a cauldron, like the wizard of the future that you are.

⚖️ “With a leveraged worker, judgment is far more important than how much time they put in or how hard they work.” – Naval

🎥 Replay: Create a Coaching Relationship with AI

If you missed my webinar last week, you can watch the replay, and use 30 copy/paste prompts to deepen your relationship with an AI thinking partner.

📰 New AI News This Week

• Anthropic released Claude Sonnet 4.6 which gives Opus-like reasoning for a fraction of the cost
• Google launched Gemini 3.1 Pro which has twice the reasoning capacity of Gemini 3
• Digital Ocean reports the biggest blocker in implementing AI agents in enterprise is the cost of inference

👓 What I’m Reading

• ​The Adolescence of Technology by Dario Amodei
• ​Inside Felix: the OpenClaw AI Earning $1,000s a Week by Nat Eliason
• ​Internal AI indexes: benchmarking AI capability against your own operations

Are you ready to grow your Ai capabilities?

Join me in the Ai Coaching Academy: