I wonder if this took into account GPU render-time fingerprinting? The minute timing differences in how you render an image are still used to identify you.

▲

magicalhippo 4 hours ago | parent | next [-]

Recently did a WebGPU project and noticed Firefox polls GPU status on a 100ms timer, so you can't really use it for such as far as I can tell.

That said, Claude figured out that while Canvas copying of a loaded JPEG was out due to fingerprint resistance, it could copy it via WebGPU, so yeah...

	▲	LoganDark 21 minutes ago \| parent [-]
		This reminds me of some fucking bullshit about how browsers silently store canvas pixel data as premultiplied and try to transparently convert everything you put in and everything you take out, but silently corrupt color values in the process. https://github.com/nagadomi/nunif/issues/32#issuecomment-153...

▲

raphlinus 3 hours ago | parent | prev [-]

If you can run compute shader code, the fingerprinting and side channel possibilities become very rich. [1] is one paper among many on this topic.

[1]: https://dl.acm.org/doi/10.1145/3634737.3656283