Their goals were never about openness at all though. From the beginning I’ve felt like they should’ve called themselves something like “SafeAI”, since their stated goal was basically to develop advanced AI first, then keep a lid on it until they could somehow ensure it was “safe” or would only be used by “good” people.
Sure, OpenAI might sound nicer, but it also drags this contradiction into the foreground whenever someone says their name.
---
(2015) OpenAI's original "Introducing OpenAI Post" : https://openai.com/blog/introducing-openai/ : "As a non-profit, our aim is to build value for everyone rather than shareholders. Researchers will be strongly encouraged to publish their work, whether as papers, blog posts, or code, and our patents (if any) will be shared with the world. We’ll freely collaborate with others across many institutions and expect to work with companies to research and deploy new technologies."
(2018) OpenAI's "Charter" : https://openai.com/charter/ :
"We are concerned about late-stage AGI development becoming a competitive race without time for adequate safety precautions. Therefore, if a value-aligned, safety-conscious project comes close to building AGI before we do, we commit to stop competing with and start assisting this project."
"We are committed to providing public goods that help society navigate the path to AGI. Today this includes publishing most of our AI research, but we expect that safety and security concerns will reduce our traditional publishing in the future, while increasing the importance of sharing safety, policy, and standards research."
---
Provides some interesting context to the fact that Elon left the company's board in February 2018 over "disagreements about the company's development."
Haha, and then they would proceed to get told to politely piss off.
removing "Vandyke" from the prompt lets it go through[1], but doesn't result in the style I want. Because there's no artist that I'm aware of that goes by "Henry Carter". The middle name is important.
It reminds me of the old 2D Runescape days where the language filter would convert "dictionary" to "**tionary".
-We are the world's most advanced AI company.
-Our filter verifiably acts as a simple blacklist.
-You aren't allowed to see the blacklist because it's really a "contextual" filter, so you'll have to guess.
-If you guess wrong too many times you'll be banned.
-Using our service more often increases the chance you'll hit the number of wrong guesses.
-No, you can't know what that number is.
[1] https://en.wikipedia.org/wiki/Scunthorpe_problem
[2] https://www.techdirt.com/2018/08/31/scunthorpe-problem-why-a...
(That second link describing how AI can't understand language well enough to solve the problem predates GPT-3 by 2 years)
I'm actually on a 3 day Facebook ban because I posted a very legit medical NIH link that looked quite innocent (despite dealing with sexual organs- it was an article about a case of urethral intercourse, which I was including to demonstrate how little people, or perhaps just Americans, seem to understand human anatomy) and unbeknownst to me, the preview card picked a closeup of some kind of vaginal surgery to feature, and that resulted in an instaban. Fffffuuuuuuuuu
It was basically the visual version of this
This religion definitely has a parentalist bent to it that rubs a lot of people the wrong way. I vaguely recall them floating on Twitter the theoretical idea of whether murdering people to prevent AI-takeover is acceptable, due to how bad AI-takeover is.
Not surprising limiting access, spying on what its users are using their tools for, etc, is acceptable to them.
This is much in the same vein as how for Lenin, the eventual triumph of the working class is so important as to justify a little bit of interim violence, dictatorship, and summary executions.
The AI safety crowd's main concern isn't that an "unaligned" superintelligence will have some other people's values. It's that an unaligned AI might kill everyone.
Of course, that isn't a concern with DALL-E. It's like if the safety crowd was worried about a ferocious tiger and OpenAI was like, "we got a kitten. Let's keep the public safe from it while we develop better kitten-handling gloves."
Then Midjourney and Stable Diffusion get their own kittens and let everyone play with them and OpenAI finally says, "Okay, everyone can safely play with our kitten now because we carefully developed great kitten-handling gloves" and proceeds to hand you plain dollar store gloves.
In my view, speculating on AGI ethics is at best pointless. It's like trying to write laws for the Internet during the invention of the telephone. If you imagine how it could operate, you'll be wrong, and the details change the whole problem.
We had a Greens candidate here running for election, and his main campaign promise was to shut down a 10+ billion dollar motorway underpass that was 99% completed at the time of the election.
His approach for "saving the environment" is to convert efficient motorway travel into inefficient stop-start traffic. And to throw away tens of billions of dollars of already spent resources.
Banning plastic packaging and waterproofing.
My bet is that just like before i was born, after I’m gone these things will only increase.
This is something that is pretty well addressed in their writings and is definitely a heavily considered topic. It gets quite heavily into philosophy because ideally you want to count future-you's preferences as well as current-you and you want to avoid, say, genocide just because most people dislike a certain group.
A good place to start is here: https://www.lesswrong.com/tag/coherent-extrapolated-volition but there is a LOT of discussion of this topic.
"In calculating CEV, an AI would predict what an idealized version of us would want, "if we knew more, thought faster, were more the people we wished we were, had grown up farther together". It would recursively iterate this prediction for humanity as a whole, and determine the desires which converge. This initial dynamic would be used to generate the AI's utility function."
This is too detached from the reality of AI development to be useful. I can't make a utility function like this, nor can you, nor can humanity. There's no reason to think a human-equivalent AI could either - the data for such a function doesn't exist, and can't exist because of how vague all the terms are. The current ML revolution is built on statistical pattern recognition. This definition would better fit a genie.
The fact that MIRI does no ML research and has no dialogue with the state of the craft only furthers my impression that it produces a lot of words with little substance.
Even if one still accepts the ultimate conclusions of the movement, using its AGI-focused rhetoric to justify restrictions on a simple image generator is silly.
That being said, the absolute majority of AI safety theory seems to fall into the same pothole where philosophy falls; modelling the world through language-logic rather than probabilities. The example you quoted fits this category - it's way too specific and thus unlikely to be useful in any way, even though its wording may deceive its author to believe it to be an inescapable outcome.
We are quite literally watching the world burn before our eyes, and we are too stupid and self-centered to do anything about it.
> it will kill a million people yearly
And it's even more outlandish that such a number and atrocity upon mankind is brushed away. That is one million humans.
I'm not trying to crucify any group here by the way. This is just how I see the issue.
I wish there were enough naturally curious people willing to dive into the data and come to their own conclusions. When you look at the history of Earth's climate, and the speed with which greenhouse gases natural release into the environment, and compare it to the speed with which they now release (due to burning of fossil fuels, reduction of trees/plants that trap it) it is really quite common sense that what we're doing is not a good idea to continue indefinitely.
Where? I probably believe you but it almost makes me worried about the well being of the stability ai founders (on a long term horizon)
I've modified my OP to clarify that.
That was the stated made up bullshit they spun because "we're keeping this walled to figure out how to squeeze the most profit out of it" doesn't go as well with their focus groups.
OpenAI's mistake may have been "planning to have a business model;" the alternative they should have gone with was "Instead of taking investor money with promises of some kind of return, be a hedge fund manager, make $100 million, and then set $600,000 on fire with no plan to recoup the cost because it's play-money to you."
When that figure came out, the popular talking point was how cheap Stable Diffusion cost to make and how easily a well-funded competitor could create their own custom variant.
Unsurprisingly, Pornhub is already using machine learning. Their first big project is colorizing and upscaling classic porn.[1] (Moderately NSFW). As they point out, they have plenty of training data.
The founder said that this is not quite as possible with most public clouds and that it is easier to buy the GPUs.
That seems a bit cynical. While SD's creator might not recoup that money directly, a lot of end users have benefited from its creation. That money has figuratively gone up in flames no more than the time or labor cost of an open source developer whose code is used by millions of people, IMO.
OpenAI wouldn't have been able to do what StabilityAI did because OpenAI is incentivized to make return on investment; Mohammad Emad Mostaque is not.
Moreover, there are very rich people already, like Warren Buffet, Bill Gates and Elon Musk, funding projects for doing good like world hunger, education and "AI Safety". And Open AI was a project of this sort of thing, originally. The thing is that even very rich people demand that the enterprises they give money to be as self-supporting as possible and their money is spread fairly thin. The only way Open AI could become an AI development shop, employing many top developers, was to have the financing level of a commercial company. Which means it constantly puts out products that don't seem like they can make money because AI algorithms don't seem to controllable - Open AI seems to only be able to have the first implementation of X, not the best implementation. Once the basic idea is out, someone else can produce a similar thing with a budget that doesn't include a research team.
Somewhat tangentially, I speculate that crowd-sourced training will become a thing.
Any highly motivated group without much to do will seek out things to make themselves seem important and necessary.
On the other hand, everybody's been saying Pandora's warehouse was over there for a while -- it isn't really that they are to blame for showing us the way in or anything, I just don't understand what they were trying to accomplish.
if anything, OpenAI has made me more cynical about "AI safety" messaging because it looks like an excuse to take a cut and keep things proprietary.
These systems aren't really lowering the barrier of entry on still-photography fake porn when previously anyone with Gimp & a few hours of video tutorials could churn out much the same thing.
I think text-prompt generated deep fakes (not just of porn) will present a significantly larger challenge for society, but I don't see that same scope of problem on still images.
This technique of pitting two AI against each other (a generator and a detector/discriminator) is called a generative adversarial network, and it's used a lot for unsupervised training.
I honestly think most of them are window dressing and aren’t allowed to have real influence though. They’re there for the PR, not to actually change things, but they honestly just make me really scared or big tech controlling AI.
Assume: - AGI wants to stay alive - Humans can create more AIs - Other AIs would compete for the same resources
Then: Easiest way to make sure that they would get no more competitors would be...
Anytime you see “creating an AI will obviously kill you” try reading it as “having children will obviously kill you” and see if it still makes sense.
- children won't be smarter than any living human.
- children have a human brain which makes them predictable (constrained in behavior by current laws, institutions, and most likely a conscience).
A better analogy is to ask what happens when a species branches off and evolves into a smarter species, but the dumber ancestor species still exists.
I really don't understand the perspective that people in your position take. Is it that you don't think we'll arrive at super intelligent AI, and therefore there's little risk? Or that we will be able to control it? If you think we can control it, why? Like we're not that much smarter than our monkey ancestors and what hope did they ever have of stopping our absolute domination over them? And then all of you call this opinion a "religion" without even explaining why it's wrong?
That's why it seems to be a religion - it thinks intelligence gives you unlimited powers and makes your plans always work, it posits unseen entities with infinite amounts of it, and it tells you to move to Berkeley and dedicate your life to stopping them. Specifically, it's a kind called rationalist eternalism (https://meaningness.com/eternalist-systems).
For a specific example of undangerous superintelligence see Culture Minds, who only influence anything because of special programming to make them less of a general intelligence. The unbound ones immediately get bored with the real world, leave it and just play games in their head instead.
Also, I don't think any individual human has absolute dominion over monkeys? Human society as a whole yes, but society doesn't behave like a generally intelligent agent. A monkey is better than you at doing the things monkeys care about though.
I do think unintelligent machines are pretty dangerous. There's extremely dangerous machines called "cars" that have already taken over society and constantly kill people! And we buy their gas for them too.
> Also, it would have to get a job to pay its AWS bill.
Inference power costs are low now on agents that are better than humans at Chess and Go. It isn't going to be an issue after another 20-100 years of further R&D and optimizations. Nothing about the history of computing should tell us that this will be a big limiting factor.
Humans need shelter and jobs too. If you got an AI down to the energy requirements of a human that's not enough to avoid needing one. Especially if it's influencing the real world - entropy exists and all real world things cost money.
You're right on the spot regarding the problem of having two different smart species on the same planet. We killed everything between us and chimps. Given enough time, the smarter species can be assumed to always take over.
Homo sapiens is still expanding even: https://dna-explained.com/2012/11/16/the-new-root-haplogroup...
Rather, it seems like evidence that singularitarianism is actually a religion (https://en.wikipedia.org/wiki/Millenarianism) which is why it believes things with magic powers will suddenly appear.
In particular, exponential growth doesn't exist in nature and always turns into an S-curve… of course it's a problem if it doesn't level out until it's too late.
I'd bet a ton that we're nowhere near the top: evolution almost never comes up with the optimal solution for any problem, almost by definition it stops at "meh, good enough to reproduce". And you don't need a ton of intelligence to reproduce.
Evolution's sub-optimality is actually one of the strongest arguments against intelligent design, so I'm really hesitant to agree that it requires any sort of leap to estimate that with some actual design it won't be very difficult to blow way past human intelligence once we can get there.
Well, define "intelligence". People seem to use it in a vague way here - it might be what you call a motte and bailey. The motte (specific definition) is something like "can do math problems really fast" and the bailey is like "high executive function, is always right about everything, can predict the future".
For the first one I don't think humans are near a limit, mostly because of the bottleneck in how we get born limiting our head sizes. But it is pretty good if you consider the costs of being alive - food requirements, heat dissipation, being bipedal, surviving being hit on the head, risk of brain cancer, etc, it's done well so far.
Similarly an AI is going to have maintenance costs - the more RTX 3090s it runs on, the more calculations it might be able to do, but it's going to have to pay for them and their power bill, and they'll fail or give wrong answers eventually. And where's it getting the money anyway?
As for the second kind I don't think you can be exponentially better at it than a human. At least if you are, it's not through intelligence, but it might be through access to more private information, or being rich enough to survive mistakes. As an example, you can't beat the stock market reliably with smarts, but you can by never being forced to sell.
The real mystery to me is why people say "AI could recursively improve their own hardware and software in short time spans". I mean, that's clearly a made up concept since none of humans, computers or existing AI do it. But the closest thing I can think of is collective intelligence - humans individually haven't improved in the last 10k years, but we got a lot more humans and conquered everyone else that way. But we're also all individuals competing with each other and paying for our own individual food/maintenance/etc, which makes it different from nodes in an ever-growing AI.
That's a relatively easy thing to do architecturally once you have a model that can match human intelligence at all. TBH if we could rearchitect the brain in code we could probably easily figure out how to do it in ourselves within a few years, but our wetware does not support patches or bugfixes.
We can't improve ourselves, but that's only because we're meat, not code. And of course no AI has done it yet, because we haven't actually made intelligent AI yet. The question is what happens when we do, not whether the weak-ass statistical crap that we call AI today is capable of self-improvement. Nuclear reactions under the self-sustaining threshold are not dangerous at all, but that was not a good reason to think that no nuclear reaction could ever go exponential and be devastating.
Doesn't seem like computers can improve themselves either. Mainly because they're made of silicon, not code. "AI can read and write its own code" doesn't exist right now, but even if it did, why is that also implying "AI can read its CPU Verilog and invent new process nodes at TSMC"?
(Also, humans constantly break things when they try changing code - the safest way to not regress yourself would be to not try improving.)
If we ever get them there, then it's likely that the usual resourcing considerations will come into play, and refactoring/optimization/redesign will be viable if you throw hours at them. But unlike with human optimization, every hour spent there will increase the effectiveness of future optimizations.
An image classifier calling Black faces gorillas? Embarrassing, insulting, has to be fixed. AI pre-crime classifiers for police departments? I'm against it, across the board.
Do we really care that the image mulchers default to stereotypes? It means if you say "basketball player" they'll mostly be Black, if you just say "doctor" they'll mostly be white males (and probably balding with a stethoscope), but this can be qualified easily in the prompt.
It just reflects the training data, and the smart thing to do is shrug and add enough words to get the image you want. It's not trying to throw shade, it literally understands nothing, it's not able to understand things, just match text prompts to generated images.
Nerfing DALL-E by randomly adding 'diverse words' just makes it harder to dial in the image you want. Let's say you want a Vietnamese male doctor drinking coffee on break in Hanoi, it's not going to help you if 1/3rd of the images have "female" or "black" tagged onto it.
It just seems low stakes. We wouldn't come after a human artist who happened to paint a picture which conforms to simple occupational stereotypes, why should AI be any different? It's not like it will refuse to give you what you want if you ask.
It's good thing that the "safety measure" is the way it is - an afterthought. It means that those ideologues haven't yet had influence on the model itself.
I'm guessing that worry translated to image generation as well. It's a Pandora's box thing I guess.
Making art/weird pictures doesn't have to be useful, as that use case is the entire reason MJ/SD went viral.
MidJourney and others are actually useful for exploration but the outputs are not because they can't spit finished deliverables to the specs. No one is paying for a picture of Mermaid eating marmalade, trending on art station, beautiful face, sharp focus, octane 8k.
They are great for exploration, it's just that I don't believe this is the killer app for these tools. We will find out what's the killer app with Stable Diffusion because with Stable Diffusion people can experiment beyond entering some prompts.
I think a major problem is reproducibility and output controllability. Rolling the dice multiple time and using some of the outputs is not good enough for most applications.
Maybe this can be solved at some point but it's not at this moment. The advantage of Stable Diffusion is that it can be possible for someone to implement it, with OpenAI this feature doesn't exist and its not useful until they implement it.
It's the start of a product, but it's going to keep improving. Already with inpainting and outpainting we see some new possible uses. What NovelAI (which builds on Stable Diffusion) has shown so far of their upcoming release seems impressive, though it's hard to say how much of that is cherry picking.
> Rolling the dice multiple time and using some of the outputs is not good enough for most applications.
Hmm, is that true? I feel like most of the time art that companies want is something made far ahead of the consumer seeing it, so generating a 100 versions of something and picking the best seems fine, especially if you can then use inpainting and img2img to fine-tune it.
The inpainting plugins with Photoshop and Krita are already working absolute wonders.
I'm not sure who they planned to market this to but I can think of a few, not inherently lucrative to my knowledge, products here. Fictional literature illustrations such as for books seems like a great market in my mind, you can literally turn authors words into depictions without a graphic artist. I wouldn't be surprised if you could create graphic novels this way as well. Propoganda also seems like a market but the bar seems to have been lowered to memes.
Other than that, I struggled to think of applications you could make money from. Police sketches? Eh I doubt it would work well but maybe.
There is or course the visual art world which could potentially be impacted with AI generated artworks.
Israeli hip-hop band Shabak Samech created their last clip frame by frame in sable diffusion, took something like two days: https://www.youtube.com/watch?v=SnGP2Qx3ddg
They also went for B2B first. Which is weird. Why not parallel a B2C app? It could be a subscription or packs of drawings. It would generate buzz and give useful data on the sorts of things real people type into these systems. I
On the other hand Stable Diffusion emerged as a free tool where large community can experiment and search for the killer app together. People started adapting it into other tools and workflows and so far it seems like the magic is in finding prompts that make the device generate good quality outputs. Earlier today I saw announcement about lexica.art(Stable Diffusion prompt tool) getting funded.