> I don't remember the details with encryption but it was basically you have to ship a breakable version for the rest of the world, and you generally sometimes ship a backdoored version.
I do remember the details: the result of Bernstein v. United States was that you have a First Amendment right to publish code because it is a speech act and so the USGOV cannot prevent you from publishing effective encryption algorithms. Will model weights be afforded the same protection? What about serving a model without publishing its weights? We shall see.
This is indeed a great quote (one of many gems from Sir Tony) but I think the context that follows it is also an essential insight:
> The first method is far more difficult. It demands the same skill,
devotion, insight, and even inspiration as the discovery of the simple
physical laws which underlie the complex phenomena of nature. It
also requires a willingness to accept objectives which are limited
by physical, logical, and technological constraints, and to accept a
compromise when conflicting objectives cannot be met. No committee
will ever do this until it is too late.
Cuckoo filters outperform bloom filters and allow dynamic insertion and deletion (unlike bloom filters, which only allow insertion). The trade off is that insertion can fail if the table is too full and then would need to expand or store those entries some other way to avoid a false negative.
The problem in this conversation is that you are equivocating between "fixing memory safety bugs" and "preventing memory safety bugs statically." When this blog post refers to "memory safety skeptics," it refers to people who think the second is not a good way to expend engineering resources, not your imagined flagrantly irresponsible engineer who is satisfied to deliver a known nonfunctional product.
I do remember the details: the result of Bernstein v. United States was that you have a First Amendment right to publish code because it is a speech act and so the USGOV cannot prevent you from publishing effective encryption algorithms. Will model weights be afforded the same protection? What about serving a model without publishing its weights? We shall see.