Comment by reallymental | Hacker News Reader

Comment by reallymental | Hacker News Reader

Parent Full thread

Both the child comments to my parent allude to the same thing. The chemical companies are under mountains of regulations, all reinforced with engineering societies (which may be just as toothless, but at least they're guaranteeing that the universities that adopt their standards, have a certain level of education their students absorb, about the dangers of what they're working on and how to make it safer.), education, societal pressure etc.

Mech engineers always overbuild things with safety factors in the 2x and 10x levels.

Why not just remove the neural/earth wires from all buildings and sockets? That'll do a whole lot of good by saving a bunch of copper, that's a tight resource I hear.

Also, screw air/sea regulation as well. We'll all just stand in the cabins, that'll increase Ryan air's failing profitability. Charge me for bringing my shoes on board too, I'm too rich anyway, skim any change off of my pocket please.

This slope is very very slippery.

This culture around rigor and safety in chemical plants came many decades after their introduction as a general class of industry. They used to just dump the waste directly into the rivers and it wasn’t even illegal.

This is a new type of technology, there is no real sense of its limits or capabilities. Regulating it appropriately or even just understanding best practices for using it is still way out by many years.

And also every country is working on it so it will get built and used for economic and military purposes, by someone.

This is just the nature of it. And yes there’s nothing at all that stops the US from doing something self destructive in terms of over regulation. I suspect this will happen eventually but Chinese models are really good.

> new type of technology [...] it will get built and used for economic and military purposes, by someone

Frank Rosenblatt, the Perceptron: 1957, planned for military purposes. Perfected image recognition came 15 years ago. I cannot understand why some people are talking about "new technologies" - all I may see is that the Military may inquire and solicit, given the hype, "Can you prepare something new for us". Otherwise, I cannot see the "novelty" about AI employment in the military now.

(I see specific areas of boost of NNs in the military - but nothing for the mainstream.)

At a very basic level if you can use frontier LLMs to do cutting edge mathematical research then you can use it for military purposes.

More specifically the jump between GPT 2.0 and GPT 3.0 with its in context learning was a step change in functionality compared to anything else that had been done with NNs, let alone MLPs.

I would be scared to my normal extreme levels if anyone in the loop trusted that.

Trust what? There’s essentially unanimous consensus in the mathematical community that LLMs are revolutionary for math research. You understand these things can and are used to generate Lean? You don’t need to “trust it” in the typical sense. It is used to generate Lean proofs which are statically checked.

Sure - that will require trained operators though, or it will make a mess in the hands of people without proper workflows.

Trained operators and workflows to do novel mathematical research? Your mental model of the world is completely wrong

Apologies for the lack of clarity. But I was thinking of potential abuses in the whole possible processes that could be impacted when the risks are not clear. You do not know of employees using bad shortcuts when unseen?

All the openings for sloppy behaviour are dangerous.

Edit:

> completely wrong

You meant "definitely" ;) ...No, it is not. It is just that perfect expression has a cost that I cannot always afford.

Good point - but I would not be able to easily define good obvious universal constraints for the technologies relevant to the submission.

Reply on news.ycombinator.com