Reading between the lines: they’re done with open source. Meta AI will be closed source like the other US models going forward.
Discussion
which means they become even more specifically trained to manipulate you
I knew that LLaMa was proprietary trash.
I never used LLaMa since 2024.
True. The new Chinese open models like Qwen 3 Coder and GLM 4.5 seem pretty good.
I had used Qwen 2.5 Coder, which is pretty nice I think. I was actually attempting to grab the 2b parameter version of DeepSeek Coder V1 (as I'd retroactively call it) for Ollama so I could use the tooling that's supported.