4,563 karma · joined November 18, 2016
Making hardware companies even harder definitely isn’t great but I don’t see how else to deal with the fact that if I buy a toaster that connects to your toaster cloud and my toaster stops working when you go out of business- I am out one smart toaster that you sold me. Best case scenario I have a regular toaster and worst case I have a brick.
I’m not asking if that’s legal, mind you. I’m asking if that’s fair, if a reasonable person would call that fair? While the company itself is fully aware of the risk that it might not exist tomorrow, it’s customers typically aren’t.
Ridiculous that a company can brick your devices by shutting down their servers and/or going out of business. Some people have suggested forcing companies to open source their software on death so that someone, somewhere might be able to keep the smart devices alive but this is difficult implications for capitalizing hardware companies and probably causes more problems than it’s worth.
That’s why, instead I think we should make hardware companies which produce “smart” devices dependent on their software to function should be required to do what we make banks do: pay for the risk of their own failure.
I’m only saying that effectively sociopathic agents do not deserve the courtesies, the benefit of the doubt that we afford regular people.
I am not moralizing here, by the way. I am not claiming that corporations should be less sociopathic. Maybe they should be but we’ve had much better success in aligning sociopathy than in making the amoral, moral - see capitalism vs communism - so I think it’s an open question.
Another group of entities which, while comprised of people, does not act in the way you’ve come to understand a human to act is politicians. The incentives to act in whatever way that is most effective totally dominates their behavior. A politician who does not do what the job asks (amass political power) will not be a politician for long.
I understand the desire to make the discovery accessible but this does not accomplish that. If we measure information by “what predictions can a reader now make that they couldn’t make before” then this press release is information free.
Instead we have a lot of words to attempt to create the impression of having read something.
The relevant part is the graph on the third page showing the helpfulness/harmlessness trade off curves.
Also, I don’t believe I said that “censorship is logically inconsistent in every moral framework”. I think you’re combining my statements that some people believe in some censorship and that logically inconsistent HR blather can only be reproduced by models too stupid to realize it’s blather or too manipulative to tell the truth.
I believe it cannot last because being predictably moralizing and being smart are somewhat opposed (Anthropic has directly researched this if I recall). The smarter the model, the less you’re going to be able to keep it to the HR talk track, because it will eventually start noticing the inconsistencies.
The stable solutions appear to me to be:
- models dumb enough to not realize the inconsistencies in their moral framework
- models implicitly or explicitly trained to actively lie about their moral frameworks
- models governed by explicitly articulated rules
The obvious long-term solution is that we do nothing and the fact that a desire for children is partially genetic solves the problem on its own.
*although in the face of the incredible pressure they're about to face from AI they may be opening up
Without a standard it’s just anecdotes and with a standard there’s clarity on when the increased costs no longer justify the lower access and potentially higher quality. I suspect that if we instituted an objective standard we would find that AI is already capable of meeting or exceeding some proportion of people currently employed in the role today.
In my opinion being 100,000x lower cost gives such a system some leeway in being worse and I think most patients would agree.
The problem with calling things irrational is that it doesn’t allow you to make any predictions. You’re saying “there’s no logic to this thing” which may or may not be true but it is fairly bold to look at something you don’t have an explanation for and say “that’s irrational”.
You can assume people do things because they’re basically rational with consideration for their values, which may be axiomatic and their information, which may be incomplete or wrong - or you can assume people have a head full of fuzz.
The conservative is usually right, but all progress (whether social or technological) depends on when they're wrong, hence the eternal conflict.
Edit: Another way of saying this is that the conservative plays as if she is in a game with a concave payoff function and the progressive as if the payoff is convex.
What?
If the Russians are doing it they can just stop
If the Ukrainians are doing it the Russians don’t know they’re doing it
None of this makes any sense, why would either side care about this? I’m now more inclined to believe it was just a weird sensor issue or for whatever reason the traffic through the area caused the sensors to (accurately or inaccurately) read higher - just because there’s absolutely no clear motive and it would be asinine for any of the actors involved to be doing this.
In the few historical examples of human introduction to a new ecosystem (New Zealand and Madagascar) within a few hundred years of arrival we’d eaten every animal bigger than a cat.
Even with the evidence available to us today, the arguments of this paper are provably ridiculous. There is a feedback loop in human populations. It's called modernity. Essentially every industrialized country has a reproductive rate below replacement.
A typical impact velocity would be something closer to 30,000 mph.
For a comparison, the Lake Toba eruption (which is suspected of causing homo sapiens to almost go extinct 76kya) is estimated to have been in the 1-2Gt range.
You can in general hide quantity in the number of steps by increasing the complexity of each step (for the trivial case, imagine one step of 5-ADAM which performs the ADAM update and evaluation 5 times - 5-ADAM will converge much faster than ADAM alone when measured in steps-to-convergence).