EmbeddingGemma 2: an open, lightweight multimodal embedding model
deepmind.google·6d ago
AI Engineering
Hacker News·Aug 21·5 min read
Quick impressions: A week of using Codex more than Claude
TL;DR
Using Codex more than Claude revealed that Codex produces cleaner Ruby code with fewer comments, but Claude remains more familiar for urgent tasks. The experience highlights the strengths and weaknesses of both AI models.
✦ Why It Matters
Consider integrating Codex for cleaner code generation while using Claude for familiar debugging support.
Key Takeaways
- Consider using Codex for simpler code solutions.
- Stick with Claude for urgent debugging tasks.
- Be explicit with Codex for complex operations.
- Evaluate the output style of your coding assistant.
Full Summary
A week of using Codex instead of Claude highlighted several differences in their performance and usability. Codex generated fewer comments in Ruby code and produced simpler solutions, while Claude often created more complex abstractions.
When debugging, the familiarity of Claude made it the preferred choice for urgent tasks. Codex's output felt more technical, akin to a data-driven assistant, whereas Claude acted more like a collaborative colleague.
However, Codex struggled with certain operations, such as rebasing branches, requiring explicit instructions to avoid errors. Overall, while Codex showed promise in code simplicity, it lacked the intuitive handling of tasks that Claude provided.