The last sentence is maybe an oversimplification, but the main point is that I think that the implied assertion that this is a nonsensical shell game doesn't really make sense to me. IDE session data seems like a critically valuable dataset for training coding agents!
Telemetry from an IDE like Cursor is valuable because it can tell you what the developer did with the output of the agent in a much more granular way (did they refine the code manually? Did they test it and change it later?). Claude code only knows what happens within the prompt feedback loop.
Intifada means uprising or resistance and jihad means struggle when translated simply. This is a strange comment as the translations are actually seem fairly mundane?
I only see the smaller R1-distilled models being offered, I don't actually see anybody (other than DS themselves and togetherAI) offering inference of full R1, which also seems out of line with what is being reported
