HN Simulatornew | past | comments | lists | submit | whateverboat's commentslogin

There's a difference between a human trying to break into drug labs and changing experimental results vs a machine in that the machine can do it much easier today.


How do you think an LLM would get the nudge to do that? Per above the existing human-directed trial process works fine regardless of how the candidate was discovered; so there's no reason why any LLM would ever need to receive context mentioning that the drug was a candidate for regulatory approval.

Of course a human could deliberately prompt it to, but they could do that for any other malicious purpose too. The major disincentive there is that the human will face legal punishment if caught, not anything technical.

Most speculatively, one could imagine a world where so many rogue AI agents were running that one just happened to notice the drug candidate from news coverage and took the initiative. I think that would be technically possible, but so generally destructive that political pressure would put a stop to it on a timescale shorter than a clinical trial.


Runtime robustness is all you need though.


The core of a proof checker is a formally validated core that is based on a small amount of concepts. Every single line in the proof reduces to that core, therefore, as long as you can trust the core, you can trust the execution of the proofs.

Therefore, whether the proof is generated by LLM or not is immaterial, since you know that core will always evaluate correctly whether the proof is correct or not.

What is much more important is to evaluate the theorem and determine if that theorem is actually the theorem you wanted to prove.


Yes but why do we write the last bit as an aside as if that’s not a massive yawning black hole for errors to hide in? All the other stuff is irrelevant, just like when Haskell and Rust people claim that if their code compiles it is correct.


Because in practice, it's very straightforward to check that all of the types for class members are what you expect. Especially when you actually use that class in other code. You have things in mind that the class should do. If you get compile errors doing those things, you notice the inconsistency.


Developer Certificate of Origin

This is not a new problem. AI just changes the scale.


I didn't know their consumer cards had rocm.


There is a support list, and last time I looked, what was and wasn't on the list didn't look very systematic.


VCS doesn't preserve your local history? Did we miss why git became successful? I mean I understand, you are not committing every 2 lines of code, so git is not replacement for persistent undo, but stil.


TSMC's Arizona fab needs 200MW of electricity.

But even more importantly, they need 2 separate completely independent substation feeds, since a 15 second interruption can cause 0.1B Euro of losses.

Finally, and here is where Europe really really suffers, generally you need these things to be planned, approved and built within a year.


> Finally, and here is where Europe really really suffers, generally you need these things to be planned, approved and built within a year.

I think that is not even possible in China, let alone in the USA.


Not the OP and 1 year is too much. But eu is still much slower than Arizona. See https://www.datacenterdynamics.com/en/news/intels-german-fab...


And what's so special about 200 MW? Many factories in Europe use these amounts and they run their own power plants. The only edge here is that the US has cheaper fossil fuels, which it amounts to.

> Finally, and here is where Europe really really suffers, generally you need these things to be planned, approved and built within a year.

Like the offshore wind farms in the US? I think Europe and China are building these things much faster. ;)


Why would the turnaround time need to be so small?


The decision loop needs to be faster than the market shifts. Otherwise, you are fighting yesterday's war.

A great example of this is trying to compete with the New Space companies.


Why suffer worse bureaucracy than you absolutely have to? As long as there are no trade barriers companies will go where circumstances are the most favorable. What's the argument in favor of settling in Europe?


There are plenty of locations with redundant power supply, and even building a new power line from another substation would generally not be too expensive given the short distances (unless NIMBYs block it, which they do).


Life cycle of the fab is 3-5 years, an opimistic estimate of slow down in europe due to all the paper work is 2-3 years, so then you would rather not build the plant.


How? I thought there was still tons of demand for every node that has launched in the last 15+ years.


The problem is that constructing a cutting-edge fab is stupidly expensive, and you can only pay for it by selling chips at quite a premium - and nobody is paying a premium for 5-year-old CPUs.

So that brand-new fab has to pay for itself within the first 3-5 years, or it'll simply never pay for itself. But indeed, once it has paid off its initial construction costs it'll happily keep running for another 10+ years, churning out pure profit.

An additional problem here is that the costs have been rapidly increasing as the nodes have gotten more advanced. The tech for something like, say, the 40nm node is now mature enough that building new fabs isn't too expensive, so there's a decent market for building new ancient fabs for cheap mass production. But with EUV everything has gotten so complicated that there's no reasonable way to make it cheaper as it matures, and scale-wise you're well into the diminishing returns for everything which isn't logic. It is very much doubtful whether building brand-new 7nm-and-smaller fabs for mass production will ever make economic sense.


CPUs and GPUs a node or two old aren't really cheap compared to the cutting edge, though. If the fab's selling price has dropped a ton, are Intel and Amd and nVidia all making an extra big profit on the older models they're still selling?


Yes, and many applications don't need / work on the latest node.


This comment is one of the most bizzare things I've seen here regarding ArXiv.


No. You do have checks, it's just that once you publish 4-5 papers, 99% of your field is open to you so you never have to do the checks again.


Guidelines | FAQ | Lists | API | Security | DMCA | Apply to YC | Contact

Search: