@naval Question, what happens when x% is generated by AI, and then that same amount is harvested and used for training? Since its good at predicting, wouldnโt that mean that some time in the future, writing material will become identical or similar?
Open Source saves the day again. Today, I cancelled my Claude pro and officially moving to:
- Questions/quick inquiries(scouting): Deepseek-v4-flash
- Code: Deepseek-v4-pro
- Planning/review: Kimi-k2.6
No more usage limits & no performance impact for my use case.
To go further, I also switched the models.
Deepseek-v4-pro for coding
Kimi-k2.6 for planning
Result: $20, and still going 3 days straight. No more โyou hit your limitโ.
Spent the last few days testing multiple agent CLI. I was using Claude Code at first, then switched to OpenCode, then Hermes, back to Claude Code and finally settled with Pi.
Best discovery and decision! No more headaches, just freedom offer by a tool that is fully customizable
@heyandras Its become a standard ๐ฅ Iโm sharpening my Data engineering skills and thanks to coolify, I can run apache Airflow, Kafka and Spark on my hardware ๐ฅน