Good start, but ‘Documents’ is the wrong, bureaucratic approach - if there are any papers they will be vague and old. The algorithm is code, and just a few characters adjusted in an obscure place can make a huge difference. I recall twitter published their old algorithm - scala code - a couple of years back, so what we need is a continuous update of that, together with some testing mechanism to check that it’s really the same as applied on their servers, without extra secret please-boss-backdoor-tweaks. Of course even better in the longer term - attract people away to the fediverse.
Meanwhile more general problem - european union doesn’t support enough people who understand such code.
Note also some good comments below the article.
In my experience working in FAANG, the docs are worth a lot more than the code. The code for the algorithm might be affected by thousands of different things, but the docs would give an overview.
It seems to me you are assuming an honest process - at least a concept of accountability to shareholders or some open governing body. In this case, one megalomaniac guy is now in charge, so to have any trust that documents are really applied to such algorithms we’d need an open, verifiable testing system - doubt it’d happen but it’s maybe worth the EU requesting that.
Good start, but ‘Documents’ is the wrong, bureaucratic approach - if there are any papers they will be vague and old. The algorithm is code, and just a few characters adjusted in an obscure place can make a huge difference. I recall twitter published their old algorithm - scala code - a couple of years back, so what we need is a continuous update of that, together with some testing mechanism to check that it’s really the same as applied on their servers, without extra secret please-boss-backdoor-tweaks. Of course even better in the longer term - attract people away to the fediverse.
Meanwhile more general problem - european union doesn’t support enough people who understand such code.
Note also some good comments below the article.
In my experience working in FAANG, the docs are worth a lot more than the code. The code for the algorithm might be affected by thousands of different things, but the docs would give an overview.
It seems to me you are assuming an honest process - at least a concept of accountability to shareholders or some open governing body. In this case, one megalomaniac guy is now in charge, so to have any trust that documents are really applied to such algorithms we’d need an open, verifiable testing system - doubt it’d happen but it’s maybe worth the EU requesting that.