Agentic data science with the Data Agent Kit
cloud.google.com·7h ago
Data Science
Simon Willison's Blog·Sep 4·5 min read
The Pelican comparison grid for Astra is pretty interesting
TL;DR
Using GPT-6 Astra, pelican images were generated at various reasoning levels and compared with outputs from GPT-5.6 models, revealing interesting insights.
✦ Why It Matters
Experiment with different reasoning levels in AI models to enhance creative outputs in your projects.
Key Takeaways
- Explore various reasoning levels in AI for creative tasks.
- Utilize comparison grids to evaluate model outputs effectively.
- Experiment with SVG generation for unique visual content.
- Analyze differences in creativity and quality across AI models.
Full Summary
Access to GPT-6 Astra allowed for the generation of SVG images depicting pelicans riding bicycles at different reasoning levels: low, medium, high, xhigh, and max. These images were then organized into a comparison grid alongside outputs from GPT-5.6 models named Sol, Terra, and Luna.
The exercise was not only entertaining but also yielded surprisingly useful results, showcasing the capabilities of the newer model. The comparison highlighted differences in image quality and creativity across the reasoning levels.
The full-quality images can be viewed in the grid, and a transcript of the prompts used for generating the pelicans is available. This exploration demonstrates the potential of advanced AI models in creative tasks.