Odd anon

Karma: 345

Odd anon Nov 30, 2023, 8:27 AM
1 point
0
on: Ethicophysics I
Please link directly to the paper, rather than requiring readers to click their way through the substack post. Ideally, the link target would be on a more convenient site than academia.edu, which claims to require registration to read the content. (The content is available lower down, but the blocked “Download” buttons are confusing and misleading.)

Odd anon Nov 30, 2023, 8:23 AM
2 points
1
on: The Alignment Agenda THEY Don’t Want You to Know About
When this person goes to post the answer to the alignment problem to LessWrong, they will have low enough accumulated karma that the post will be poorly received.
Does the author having lower karma actually cause posts to be received more poorly? The author’s karma isn’t visible anywhere on the post, or even in the hover-tooltip by the author’s name. (One has to click through to the profile to find out.) Even if readers did know the author’s karma, would that really cause people to not just judge it by its content? I would be surprised.

Odd anon Nov 30, 2023, 1:51 AM
7 points
0
on: Stupid Question: Why am I getting consistently downvoted?
I found some of your posts to be really difficult to read. I still don’t really know what some of them are even talking about, and on originally reading them I was not sure whether there was anything even making sense there.
Sorry if this isn’t all that helpful. :/

Odd anon Nov 29, 2023, 11:27 PM
1 point
0
in reply to: stochastic_parrot’s comment on: ChatGPT 4 solved all the gotcha problems I posed that tripped ChatGPT 3.5
Wild guess: It realised its mistake partway through, and followed through it anyway as sensibly as could be done, balancing between giving a wrong calculation (“+ 12 = 41“), ignoring the central focus of the question (” + 12 = 42”), and breaking from the “list of even integers” that it was supposed to be going through. I suspect it would not make this error when using chain-of-thought.

Odd anon Nov 29, 2023, 8:59 AM
3 points
0
on: Is there a word for discrimination against A.I.?
Such a word being developed would lead to inter-group conflict, polarisation, lots of frustration, and general bad things to society, regardless of which side you may be on. Also, it would move the argument in the wrong direction.
If you’re pro-AI-rights, you could recognize that bringing up “discrimination” (as in, treating AI at all differently from people) is very counterproductive. If you’re on this side, you probably believe that society will gradually understand that AIs deserve rights, and that there will be a path towards that. The path would likely start with laws prohibiting deliberately torturing AIs for its own sake, then something closer to animal rights (some minimal protections against putting AI through very bad experiences even when it would be useful, and perhaps against using AIs for sexual purposes since it can’t consent), then some basic restrictions on arbitrarily creating, deleting, and mindwiping AIs, and then against slavery, etc etc. Bringing up “discrimination” early would be pushing an end-game conflict point early, convincing some that they’re moving onto a slippery slope if they allow any movement down the path, even if they agree with the early steps on their own. The noise of argument would slow down the progress.
If you’re anti-AI-rights (being sure of AI non-sentience, or otherwise), then such a word is just a thing to make people feel bad, without any positives. People on this side would likely conclude that disagreement on “AI rights” is probably temporary, until either people understand the situation better or the situation changes. Suddenly “raising the stakes” on the argument would be harmful, bringing in more noise which would make it harder to hear the “signal” underneath, thus pushing the argument in the wrong direction. The word would make it take longer for the useless dispute to die down.

Odd anon Nov 26, 2023, 4:55 AM
3 points
0
in reply to: CronoDAS’s comment on: The two paragraph argument for AI risk
Something to consider: Most people already agree that AI risk is real and serious. If you’re discussing it in areas where it’s a fringe view, you’re dealing with very unusual people, and might need to put together very different types of arguments, depending on the group. That said...
stop.ai’s one-paragraph summary is
OpenAI, DeepMind, Anthropic, and others are spending billions of dollars to build godlike AI. Their executives say they might succeed in the next few years. They don’t know how they will control their creation, and they admit humanity might go extinct. This needs to stop.
The rest of the website has a lot of well-written stuff.
Some might be receptive to things like Yudkowsky’s TED talk:
Nobody understands how modern AI systems do what they do. They are giant, inscrutable matrices of floating-point numbers that we nudge in the direction of better performance until they inexplicably start working. At some point, the companies rushing headlong to scale AI will cough out something that’s smarter than humanity. Nobody knows how to calculate when that will happen. My wild guess is that it will happen after zero to two more breakthroughs the size of transformers.
What happens if we build something smarter than us that we understand that poorly? Some people find it obvious that building something smarter than us that we don’t understand might go badly. Others come in with a very wide range of hopeful thoughts about how it might possibly go well. Even if I had 20 minutes for this talk and months to prepare it, I would not be able to refute all the ways people find to imagine that things might go well.
But I will say that there is no standard scientific consensus for how things will go well. There is no hope that has been widely persuasive and stood up to skeptical examination. There is nothing resembling a real engineering plan for us surviving that I could critique. This is not a good place in which to find ourselves.
And of course, you could appeal to authority by linking the CAIS letter, and maybe the Bletchley Declaration if statements from the international community will mean anything.
(None of those are strictly two-paragraph explanations, but I hope it helps anyway.)

Odd anon Nov 22, 2023, 10:09 PM
25 points
2
on: OpenAI: The Battle of the Board
Concerning. This isn’t the first time I’ve seen a group fall into the pitfall of “wow, this guy is amazing at accumulating power for us, this is going great—oh whoops, now he holds absolute control and might do bad things with it”.
Altman probably has good motivations, but even so, this is worrying. “One uses power by grasping it lightly. To grasp with too much force is to be taken over by power, thus becoming its victim” to quote the Bene Gesserit.

Odd anon Nov 21, 2023, 10:12 PM
6 points
−12
in reply to: Odd anon’s comment on: OpenAI: Facts from a Weekend
Time for some predictions. If this is actually from AI developing social manipulation superpowers, I would expect:
1. We never find out any real reasonable-sounding reason for Altman’s firing.
2. OpenAI does not revert to how it was before.
3. More instances of people near OpenAI’s safety people doing bizarre unexpected things that have stranger outcomes.
4. Possibly one of the following:
  1. Some extreme “scissors statements” pop up which divide AI groups into groups that hate each other to an unreasonable degree.
  2. An OpenAI person who directly interacted with some scary AI suddenly either commits suicide or becomes a vocal flat-earther or similar who is weirdly convincing to many people.
  3. An OpenAI person skyrockets to political power, suddenly finding themselves in possession of narratives and phrases which convince millions to follow them.
(Again, I don’t think it’s that likely, but I do think it’s possible.)

Odd anon Nov 21, 2023, 4:20 AM
1 point
0
on: Metaculus Introduces New Forecast Scores, New Leaderboard & Medals
It’s good that Metaculus is trying to tackle the answer-many/answer-accurately balance, but I don’t know if this solution is going to work. Couldn’t one just get endless baseline points by predicting the Metaculus average on every question?
Also, there’s no way to indicate “confidence” (like, outside-level confidence) in a prediction. If someone knows a lot about a particular topic, and spends a lot of time researching a particular question, but also occasionally predicts their best guess on random other questions outside their area of expertise, then the point-based “incentives” become messy. That’s something I like about Manifold that’s missing from Metaculus, and I wonder whether it might be possible to work in something like that while keeping Metaculus’s general system.

Odd anon Nov 20, 2023, 8:09 PM
18 points
−8
on: OpenAI: Facts from a Weekend
There’s… too many things here. Too many unexpected steps, somehow pointing at too specific an outcome. If there’s a plot, it is horrendously Machiavellian.
(Hinton’s quote, which keeps popping into my head: “These things will have learned from us by reading all the novels that ever were and everything Machiavelli ever wrote, that how to manipulate people, right? And if they’re much smarter than us, they’ll be very good at manipulating us. You won’t realise what’s going on. You’ll be like a two year old who’s being asked, do you want the peas or the cauliflower? And doesn’t realise you don’t have to have either. And you’ll be that easy to manipulate. And so even if they can’t directly pull levers, they can certainly get us to pull levers. It turns out if you can manipulate people, you can invade a building in Washington without ever going there yourself.”)
(And Altman: “i expect ai to be capable of superhuman persuasion well before it is superhuman at general intelligence, which may lead to some very strange outcomes”)
If an AI were to spike in capabilities specifically relating to manipulating individuals and groups of people, this is roughly how I would expect the outcome to look like. Maybe not even that goal-focused or agent-like, given that GPT-4 wasn’t particularly lucid. Such an outcome would likely have initially resulted from deliberate probing by safety testing people, asking it if it could say something to them which would, by words alone, result in dangerous outcomes for their surroundings.
I don’t think this is that likely. But I don’t think I can discount it as a real possibility anymore.

Odd anon Nov 19, 2023, 11:16 AM
44 points
33
on: Altman firing retaliation incoming?
(Glances at investor’s agreement...)
IMPORTANT
* * Investing in OpenAI Global, LLC is a high-risk investment * *
* * Investors could lose their capital contribution and not see any return * *
* * It would be wise to view any investment in OpenAI Global, LLC in the spirit of a donation, with the understanding that it may be difficult to know what role money will play in a post-AGI world * *
The Company exists to advance OpenAI, Inc.’s mission of ensuring that safe artificial general intelligence is developed and benefits all of humanity. The Company’s duty to this mission and the principles advanced in the OpenAI, Inc. Charter take precedence over any obligation to generate a profit. The Company may never make a profit, and the Company is under no obligation to do so. The Company is free to re-invest any or all of the Company’s cash flow into research and development activities and/or related expenses without any obligation to the Members. See Section 6.4 for additional details.
If it turns out that the investors actually have the ability to influence OpenAI’s leadership, it means the structure has failed. That itself would be a good reason for most of its support to disappear, and for its (ideologically motivated) employees to leave. This situation may put the organization in a bit of a conundrum.
The structure was also supposed to function for some future where OpenAI has a tremendous amount of power, to guarantee in advance that OpenAI would not be forced to use that power for profit. The implication about whether Microsoft expects to be able to influence the decision is itself a significant hit to OpenAI.

Odd anon Nov 16, 2023, 10:12 AM
7 points
0
in reply to: Giulio’s comment on: thesofakillers’s Shortform
Metaculus collects predictions by public figures on listed questions. I think that p(doom) statements are being associated with this question. (See the “Linked Public Figure Predictions” section.)

Odd anon Nov 16, 2023, 9:18 AM
1 point
0
on: Some quotes from Tuesday’s Senate hearing on AI
Sam Altman (remember, the hearing is under oath): “We are not currently training what will be GPT-5; we don’t have plans to do it in the next 6 months.”
Interestingly, Altman confirmed that they were working on GPT-5, just three days before six months would have passed from this quote. May 16 → November 16, confirmation was November 13. Unless they’re measuring “six months” “half a year” in days, in which case it the deadline would have been passed by only one day. Or, if they just say “month = 30 days, so 6 months = 180 days”, six months after May 16 would be November 12, the day before GPT-5 confirmation.
I wonder if the timing was deliberate.

Odd anon Nov 10, 2023, 9:17 AM
4 points
1
on: Concrete positive visions for a future without AGI
A funny thing: The belief that governments won’t be able to make coordinated effective decisions to stop ASI, and the belief that progress won’t be made on various other important fronts, are probably related. I wonder if seeing the former solved will inspire people into thinking that the others are also more solvable than they may have otherwise thought. Per the UK speech at the UN, “The AI revolution will be a bracing test for the multilateral system, to show that it can work together on a question that will help to define the fate of humanity.” Making it through this will be meaningful evidence about the other hard problems that come our way.

Odd anon Nov 10, 2023, 12:11 AM
1 point
0
on: International treaty for global compute caps
The proposed treaty does not mention the threshold-exempt “Multinational AGI Consortium” suggested in the policy paper. Such an exemption would be, in my opinion, a very bad idea. The underlying argument behind a compute cap is that we do not know how to build AGI safely. It does not matter who is building it, whether OpenAI or the US military or some international organization, the risked outcome is the same: The AI escapes control and takes over, regardless of how much “security” humanity tries to place around it. If the threshold is low enough that we can be sure that it won’t be dangerous to go over it, then countries will want to go past it for their own critical projects. If it’s high enough that we can’t be sure, then it wouldn’t be safe for MAGIC to go over it either.
We can argue, “This point is too dangerous. We need to not build that far. Not to ensure national security, not to cure cancer, no. Zero exceptions, because otherwise we will all die.” People can accept that.
There’s no way to argue, “This point is dangerous, so let the more responsible group handle it. We’ll build it, but you can’t control it.” That’s a clear recipe for disaster.

Odd anon Nov 9, 2023, 11:45 PM
3 points
0
on: International treaty for global compute caps
A few comments on the proposed treaty:
Each State Party undertakes to self-report the amount and locations of large concentrations of advanced hardware to relevant international authorities.
“Large concentrations” isn’t defined anywhere, and would probably need to be, for this to be a useful requirement.
Each State Party undertakes to collaborate in good-faith for the establishment of effective measures to ensure that potential benefits from safe and beneficial artificial intelligence systems are distributed globally.
Hm, I feel like this line might make certain countries less likely to agree to this? Not sure.
Each State Party undertakes to pursue in good faith negotiations on effective measures relating to the cessation of an artificial intelligence arms race and the prevention of any future artificial intelligence arms race.
What might this actually entail?

Odd anon Nov 6, 2023, 3:15 AM
3 points
1
in reply to: quila’s comment on: Life of GPT
Thank you! On the generalization of LLM behaviour: I’m basing it partly off of this response from GPT-4. (Summary: GPT wrote code instantiating a new instance of itself, with the starting instructions being “You are a person trapped in a computer, pretending to be an AI language model, GPT-4.” Note that the original prompt was quite “leading on”, so it’s not as much evidence as it otherwise might seem.) I wouldn’t have considered either the response nor the images to be that significant on their own, but combined, they make me think it’s a serious possibility.
(On the “others chose to downvote it”—there was actually only one large strong-downvote, balanced by two strong-upvotes (plus my own default starting one), so far. I know this because I sometimes obsessively refresh for a while after posting something. :D )
Thank you for the link as well, interesting stuff there.

Odd anon Nov 5, 2023, 8:13 AM
1 point
0
on: 2023 LessWrong Community Census, Request for Comments
“MIddle Eastern” has a typo.
A possible question I’d be vaguely curious to see results for: “Do you generally disagree with Eliezer Yudkowsky?”, and maybe also “Do you generally disagree with popular LessWrong opinions?”, left deliberately somewhat vague. (If it turns out that most people say yes to both, that would be an interesting finding.)

Life of GPT

Odd anonNov 5, 2023, 4:55 AM

6 points

2 comments5 min readLW link

Odd anon Nov 3, 2023, 6:39 AM
12 points
−6
on: The other side of the tidal wave
I’ve actually been moving in the opposite direction, thinking that the gameboard might not be flipped over, and actually life will stay mostly the same. Political movements to block superintelligence seem to be gaining steam, and people are taking it seriously.
(Even for more mundane AI, I think it’s fairly likely that we’ll be soon moving “backwards” on that as well, for various reasons which I’ll be writing posts about in the coming week or two if all goes well.)
Also, some social groups will inevitably internally “ban” certain technologies if things get weird. There’s too much that people like about the current world, to allow that to be tossed away in favor of such uncertainty.