Potential Prompt
The author discovered that using specific system instructions can significantly improve the performance of the Gemini 3 AI model, especially for technical tasks like 3D graphics and coding.
Why it matters
This discovery highlights the importance of prompt engineering and the significant impact system instructions can have on the performance of AI models like Gemini 3.
Key Points
- 1Switching on the system instructions in Gemini 3 led to better performance on technical tasks
- 2The model's reasoning style changes with the system prompt, becoming more concise and condensed
- 3Disabling bullet points in the system prompt improved the model's outputs, reducing hallucinations and errors
- 4The effect was observed across different tasks like coding, shader programming, and using lesser-known libraries
Details
The author was experimenting with the Gemini 3 AI model and found that using specific system instructions could dramatically improve its performance, especially on technical tasks like 3D graphics and coding. They noticed that the model's reasoning style changed significantly when the system prompt was enabled, becoming more concise and condensed, almost like planning in a journal. This led to better outputs, with the model avoiding hallucinations and math errors that were present when the system prompt was disabled. The author tested this by writing
No comments yet
Be the first to comment