Cone cells have fairly peaked frequency responses, and due to random projections caused by the cascading, this is sufficient to fully reconstruct the signal, i.e. the converse of the statement is true, you can retain 99.9% of the perceptual information using 3 sensors, all you need is to perceive in time and some randomness in the sensor placement.
No it does not. You can verify the data has not been modified given the HTPS even if it has passed through arbitrary untrusted process in the middle. Think about the ISP being outside the SGX enclave and having the same ability as Relay.
The example you provide on the page: "discover what you should name your kid if you want them to be a US senator" are you actually going to show causal estimates? Names are very endogenous. If you are looking at the likelihood of being a US senator conditioning on someones name, this is very different from the causal effect of the name.
the first authors twitter account https://twitter.com/polynoamial/