https://factorio.com/blog/post/fff-146

The problem is that history has shown that companies really do not want to do that. They will happily pay to maintain an inferior interface and go out of their way to disrupt third party clients.

Sure, that optimization interacts badly with the optimization that removes the infinite loop. But half the point of UB is to avoid needing to deal with such interactions, because they are defined out of existence.

In theory, there is also an annual wait limit to free dumps, but they can't actually enforce it.

If we look at his stated reason for the attack [0], it is hard to reconcile post 9/11 history with a claim of his victory.

[0] https://web.archive.org/web/20040615081002/http://observer.g...

The history of automation is filled with cases where the result has been lower quality goods at lower prices.

By itself, that does not exclude the possibility of dark matter having other interactions which do not interfere with our detectors.

If a problem is not able to influence anything, even in theory, then by definition, it cannot possible influence any testable predictions we have.

Add to that the potential for unknown and mis-moddeled manufacturing defects, and manipulated electrical voltages and signals, random particle interactions, and there is a lot of room for known physics to break a verilog level proof of correctness.

Sure a hardware or model bug would render your proof non-applicable, but that is already the case for the existing proofs.

The bigger problem is simply that hardware designers do not care about timing side channels. Even if you did accurately model the timing behavior of a modern processor, you would just discover that trying to write software free from timing side channels is a practical impossibility.

> And then you'd need assurance that the Verilog is faithfully transcribed in the silicon, which is a can of worms in itself.

You would also need to prove that our model of physics accurately describes how that silicon would behave, and the the environment around the silicon is within the physical parameters you modeled...

Otherwise, you are relying on obscurity, and will lose as soon as you become interesting enough to matter.

You will also break non-AI machine processing use cases. That isn't just accessibility, it is things like search.

> Any excess money at the end of the year should be moved to the pool for the next year, not as a bonus on CEOs' 100m+ salary

Under the ACA, excess money in the pool must be rebated to policy holders. In practice, this worked for a few years, but eventually insurance companies ended up increasing cost in order to increase the absolute amount they were allowed to keep. Maybe this is still good (more claims approved), but it is counter productive to the goal of reducing healthcare costs.

If you are Amazon, this is essentially just the multi-armed bandit problem, at it is not particularly difficult. You show the no-name product/vendor to enough people to get data for it.

[0] Funded by them, and reported on their website

[1] Their sample had a known quantity of microplastics, of which they had a 75% detection rate for their control sample.

With a strong enough regulatory framework, we probably could suppress unlisenced providers and non-watermarking local harnesses to the point where bypassing them would take some technical know-how. However, I do not think there is anywhere near the political will for such a regime.

The only practical use I can see for this in the world we actually live in is to prevent model collapse. Most people using AI don't care if people training future AI ignore them, so would have no incentive to switch to providers that do not watermark. Of course, this disencetivises all if the pro-social applications of this technology, and risks giving the big providers a monopoly on "known human" data, which has serious antitrust implications.

[0] Note that the watermark is not inherent to the model itself, but rather how the model is run. So this teqnique cannot be used by people providing open-weight models. It would need to be used by those actually running the models.

The models "natural" output is the result of a series of random numbers. The watermark works by biassing that series towards a different series of numbers. Assuming that second series is cryptographically secure psuedo-random, the even distinguishing the biased sequence from true random would be impossible with compromising the key or prng.

As an extreme, suppose your prompt was public, and the model seeded its PRNG with a secret key instead of a genuine random seed. Such an output is not meaningfully different from one based on a true RNG, but can be trivially fingerprinted by someone who knows the keys.

In practice, I am doubtful they have a scheme that is both practically useful and cryptographically secure. However, there is a lot of room below cryptographically secure that is still just as good for all other purposes.