Viewing profile — ZihangZ

Small caution from using agents here: the useful chart is the one generated from code, tests, or traces, not the one the model draws from its own explanation. I've had models produ…

+1 to the CI/isolation point. That is the part that makes these setups work for me too: make the failure cheap to reproduce, make stderr visible, make the agent rerun the same comm…

[dead]

Yeah, LAN is the line for me. USB-side sshd is a weird dev leftover; LAN means it’s now in the home threat model.

Built a ROS 2 cycling helmet. IMU + GPS, but we didn't fuse them, on purpose. Speed is just haversine between GPS fixes. IMU only does turn detection and crash/fall. No EKF. Under …

Yeah, this is pretty common once a device has any real DSP in it. There's usually some stripped-down Linux on an ARM SoC underneath, and the vendor BSP just happens to ship with ss…

[dead]

I don't think SSH vs OpenTofu is the core issue here. For agents, declarative plans are still valuable because they are reviewable. The interesting question is whether exe.dev chan…

[dead]

This matches my experience. When I write the boring glue code myself, I get a map of the project in my head. When I let an agent write too much of the structure, the code may work,…

This matches what I've seen too — the hallucination gets much worse when the loop has no external verifier. "Does this board work?" has no ground truth inside the model, so it defa…

[dead]