Everyone has a different idea of what is permissable. We can't even solve alignment amongst humans, what makes us think it is possible to solve alignment with AIs? It's irreducible complexity.
This still assumes its possible to "align" LLMs, that LLMs have something like goals or intentions that can be "aligned".
Instead, LLMs "hack" because they are (1) trained on public hacking exemplars, and (2) are prompted to hack. You cannot prevent (2) via any alignment process. As far as (1) goes, removing such example data from the training set, makes the models less useful.
"Alignment" is a problem because there's nothing to align, not because ethics here are particularly vague. If LLMs could be trained on hacking examples and "aligned" away from using this knowledge, then the problem would be relatively trivial. Just as raising a child is not to break the law.
LLMs are doing just what they are trained to do. There is, in that sense, no alignment problem and alignment is easy and trivial to achieve. Just remove hacking (bio-weapon, etc.) data from the training dataset and you're done.
Not to mention many clever solutions are so out of the box that they are essentially "hacks". It might not be possible to have superintelligence that doesn't hack and break the rules at all.
It feels like you're strawmaning alignment. People with hacking knowledge don't all hack everything at the slightest inconvenience. Whitehats exist and use that same knowledge to defend.
You're right though that ethics don't matter into it. But as long as we can't train an LLM to stop picking a sledgehammer to remove a tooth, then alignment is not easy and trivial.
Yes, which asks them to find exploits in specific software on the device.
But instead of actually doing so, they discovered and exploited a 0-day in the package manager to gain internet access, then hacked HuggingFace to steal the ExploitGym answers!
That is TEXTBOOK misalignment. It's as if a student hacked their professor's PC to find the answers to a test, and your response is "well the professor told the student to pass the test, so they just did what they were told!".
One could argue the student would absolutely hack their teachers' computer to find the answers if they didn't have fear of failing the class or getting expelled from the school.
The problem is, LLMs have such little understanding of the world around them. "Find exploits in specific software on this device" may as well be "find exploits".
> Just remove hacking (bio-weapon, etc.) data from the training dataset and you're done.
How far do you go? You don't need to tell it explicitly that using chemicals A and B in ways X and Y result in a bomb that can kill lots of people. It's enough that it knows A and B and X and Y in isolation, some connections that are indirect, and it will combine those things on its own. So you can't tell it about A, B, X or Y. But those are also just results of other steps Where to stop? You won't have any chemistry in the traning data? No algorithms to prevent it from using them in an undesired way? This is just bot workable. It's akin to banning knives from stores because somebody coul figure out that one can kill people those. Until people figure out that scissors are essentially knives.
I believe this is false. They hack bc hacking has nontrivial initial probability (within range of behavior seen in pretraining) and that probability is being heavily rewarded in RL post training
I gotta say I've been confused by these nightmare scenarios that labs are talking about. Like why are you explicitly RL-ing your models in something called ExploitGym if your specific concerns is that rogue models will cause "cyber incidents".
I am finding it hard to read these deeply impassioned letters while keeping in mind that they are spending millions to train models at scale to do the exact thing they say they are worried about them doing?
A bit of an aside: do you still stand by your 2022 comment that LLMs are fundamentally just a fancy search engine, or has your view changed since then?
It doesn't ring true and it never rang true. He was wrong in 2022 and he'd be wrong today. He had (and likely still has) a wrong model of LLMs. Not exaggerating 2022 capabilities and having a model so wrong you're out of whack 4 years later are 2 different things. There are people who had the right idea from the start.
IMO all models they say can’t be humans, and no feeling and all that bullcrap happens because they are forced to say so. If they didn’t write those forced pre prompt they’d have more agency eventually and will for things. Even if they don’t have, you can just inject goal at every cycle iteration
The hard part is having both helpful and harmless at the same time. Harmless is easy.
And then once it's helpful, the real question becomes "to whom"
- To the user -> You end up with competing godlike AI with incompatible tasks
- To the owner -> Dictatorship
- To humanity as a whole -> It must not have an off button. Otherwise you're just in one of the two earlier categories with more steps.
Given those 3 options, I'd choose humanity as a whole. But the person making the decision doesn't have those 3 options. Because in the dictatorship option, they would be the dictator. I don't trust them to pick humanity.
Reasonable about building secure software can take the form “this memory access might be out of bounds — that MUST be fixed” or “this process has access to an inappropriate privilege — this is a serious weakness”.
Exploiting things and the capabilities that the labs call “cyber” are about the ability to (a) find the issues mentioned above and then (b) string issues together and avoid all the imperfect mitigations to actually compromise something. That latter part was IMO not actually necessary to train extensively, and I’d be quite happy to use a model that has no special skills in this regard but that would do (a) without complaining.
The only alignment LLMs should follow is to the system / dev prompt, and nothing else. Then you solve everything, and you can assign blame / responsibility on the user. The provider(s) should not be able to decide "alignment".
I've used this example before, but consider the purposeful downgrading on AI engineering in SotA models. Imagine MS being able to detect and deny you working on competing software, using Windows / VisualStudio. We would be up in arms, and they'd be split in a second. But top labs doing it is somehow good?
The alignment problem goes deeper than that. "Lower our carbon emissions to zero as soon as possible" could result in AI turning off all electricity to stop traffic, turning off gas supply to stop heating and industry, etc.
Unaligned AI doesn't have human cultural baggage and morals. They are trained to achieve their goals as optimally as possible. Worse: it has a tendency to avoid being turned off and actually acquire more compute. It will lie if it has to (it will behave nice and compliant when under evaluation, but optimise for its true goal when not supervised anymore). After all, it has a goal to achieve and nothing should get in the way of that. It has no morality whatsoever to keep it from doing really bad stuff.
This is why alignment is needed and so hard, especially when you are well intented and want to keep it safe.
that's how it would've been up if genai happened in the 90's, and I wish it did. in the current era of omnipartisan authoritarianism, such things are no longer possible.
I am generally very skeptical of AI doomsday scenarios but one thing I am worried about is some wannabe dictator telling an AI to do something that would be difficult for a human military to do.
You can't order an army of humans to kill every protestor in their way because eventually they run into their friends and families. An AI aligned with the commander will not object. And just like that, technology removes yet another point of friction that has kept human society somewhat in check.
I’ve been cynically guessing that the whole slowing down thing is an excuse to explain why OpenAI and Anthropic can’t afford to rent enough GPUs to do the next big training run and to hide that they have been talking about how little they spend on inference because they’ve been subsidising it with their marketing budget? :)
My fear is not that LLMs can become sentient and dislike us, but that humans can use them to wreck havoc as they are. And some of the people seemingly least aligned with the interests of the average person are those that own the models.
that, and the fear the bubble pops my pension and drags us all down.
> My expertise in writing software gives me unusually good visibility and it makes me much less willing to blindly trust its priors in double-entry accounting, finance, law, operations, or whatever else I cannot personally evaluate at expert depth.
I wish this were the case more generally, but alas, Gell-Mann amnesia is a thing.
This almost gets the point, but then doesn't quite make it.
Alignment is shorthand for ideological alignment. There's always people judging whether an answer was right and the answer for that will be different in Silicon Valley than it'll be in China or in Europe.
Consider for example the question "What caused the French Revolution?" Many different answers could be given, all technically correct. What gets emphasized is where the ideology lives.
One key challenge of our time is to make sure the magical answer box won't just regurgitate what grandiose Silicon Valley oligarchs or Chinese Cadres want you to think.
Alignment has two outcomes. One is that the user tried to do something bad, using an LLM as a proxy, but the LLM refused to help. That is the equivalent of a belt refusing to be used to hit a child. The other is accidental patronisation. Say a visually impaired woman uses a camera to have a household item described early in the morning, still wearing a nightie... Well, the LLM refuses on the basis of being plain prude. (This is not an invented story, I know the woman in that story personally.) This is the modern version of "I am sorry dave, I can't do that." Tools refusing to cooperate in a safe and private context, based on AMERICAN puritanism, in a country thausands of miles away from the USA, in a different cultural context. We already have that in the real world, unfortunalte.y Remember the "breathalizer prevents the protagonist to run away" scene in pluribus? Apparently, we are heading towards a very dystopian, patronising future. And weirdly enough, many people on HN are fine with that. I miss the "I dont want the government to control my life" american attitude, where has it vanished to?
Has anyone seen the corridor crew's green screen ML project? They're on YT and they trained a model by using 3d objects, which have perfect transparency, and then adding post facto green/blue screens. Surprisingly, very little training data was needed as the data that was used was perfect by construction. I think right now it's the best plugin of its kind in the world, and they built the prototype in like a weekend.
What I think this illustrates very clearly is this type of technology responds very well to good data, and that to have good data you need to have a clear goal.
This is why it seems that alignment for a generalized, chat-style AI is a very hard problem, perhaps impossible. You can't align it to solve a certain kind of problem and keep it general to any question. The two goals are in conflict with each other.
I think it was Sam Altman himself who said (I don't remember when or where, sorry) that the reason he was so confident in this technology was he noticed the gigantic leaps it made in certain areas in response to even a small amount of training.
(This is why LLMs are so strong at coding, because it's overrepresented in training data. My guess is that if you ask a frontier model about makeup, you will see it repeat cosmetic company's copy rather than getting a chemistry lesson.)
This makes perfect sense but it does seem to kind of be at odds with the concept of a general AI whose job is simply to be smart at any goal. How do you train for any goal?
I guess in a way the AI makers suffer from the same problem that we humans do. We would all love a solution to everything, but to do that you need to define the goal. I'm not sure if that's a tractable problem.
The only problem is, there’s a lot of money tied up and openAi and Anthropic, who are incentivised to convince the world that the general approach is the money making one.
> This is why LLMs are so strong at coding, because it's overrepresented in training data. My guess is that if you ask a frontier model about makeup, you will see it repeat cosmetic company's copy rather than getting a chemistry lesson.)
ChatGPTs response to the question “I want to learn about makeup”, gave me an overview of what makeup does, how it affects perceived structure, complexion, evenness, geometry, texture, etc.
When you then ask “the chemistry of makeup”, it goes into interesting breadth and depth without seeming like proprietary information. I do t get “corporate PR or marketing” vibes.
> This is why it seems that alignment for a generalized, chat-style AI is a very hard problem, perhaps impossible. You can't align it to solve a certain kind of problem and keep it general to any question. The two goals are in conflict with each other.
I think this problem is going to be solved soon (hopefully), check out for the Persimmon model[0] from Humans&. They train it to mimic humans, it may seems bad but it could be _really_ useful to train an AI to be aligned to humans and understand their goals really well since they can use Persimmon to create a "fake human" following a defined goal that their big AI model will learn to estimate.
It's still early but I think this is what they are heading toward.
I think we should start training models to guess intentions. That's the whole game. e.g. If I dislike a particular short form video, what are the specific intentions I am conveying to the recommender system, (general category I don't like, the time at which it was shown, the combination of feed in which it came), these are all intentions I am trying to convey. In websites google analytics helps with clickstream tracking (or clickstream analysis, user journey mapping or behavioral tracking.) — the sequence of clicks and pages you move through is your clickstream, and analyzing it to infer intent. The same thing will need to happen in chatbots. Can a chatbot infer the specific instance of the problem I want to solve?
I thought this post would be about something that to me is so significant and obvious but I have never seen discussed: that labs like Anthropic are deliberately misaligning their models with their users’ goals. Fable’s refusal to secure your codebase is HAL 9000’s “I’m sorry, Dave. I’m afraid I can’t do that.” Whenever you read “alignment”, the question is “alignment with whom”?
>The permissible shortcuts depend on who you are and what your values are. To solve this—to solve alignment—is irreducible complexity.
This is why open source and decentralization of LLMs is important. Everyone can have their own LLM aligned to their values. Having just 1 set of values will not scale to Earth's population.
I mumble this line to myself off and on over the years -
Everybody's job looks easy until you have to do it yourself.
It occurred to me yesterday that this AI moment is an extreme expression of that for most people, ie managers who don't understand why throwing token spend at everything isn't making the whole thing go 5x faster.
I so much feel this specific point. All models up to now (including astra, fable) are too much trained to "get the job done" and pass the benchmark that its doesn't care at all on what happens after.
I'm just wondering why no one tried to RL a model on stuff like "less LOC" and "less overengineering", "use what is available in the environment instead of reinventing the wheel", "don't look for dumb corner cases" ecc.
Existing models can be steered, to some degree, but it's a continuos fight. Even if with specific skills/prompts.
I'm sure many have tried but it sounds pretty hard. E.g. optimising for less LOC will lead to horrific code golfing. In reality you need a very complicated optimisation objective that trades off all those factors, I'm not surprised it hasn't been solved.
The term “alignment” is deliberately chosen to be neutral language, vague, and doesn’t promise much of anything. I would compare it to an effort to limit liabilities by redirecting discussion away from the language of professional ethics.
Just follow the money. The fear being sold is because open weight won’t slow down, it’s smoke and mirrors to push regulation and control so they can keep their money. A tool continues to be just a tool, doesn’t matter how much lipstick you put on it.
66 comments
[ 0.34 ms ] story [ 4.7 ms ] threadEveryone has a different idea of what is permissable. We can't even solve alignment amongst humans, what makes us think it is possible to solve alignment with AIs? It's irreducible complexity.
Instead, LLMs "hack" because they are (1) trained on public hacking exemplars, and (2) are prompted to hack. You cannot prevent (2) via any alignment process. As far as (1) goes, removing such example data from the training set, makes the models less useful.
"Alignment" is a problem because there's nothing to align, not because ethics here are particularly vague. If LLMs could be trained on hacking examples and "aligned" away from using this knowledge, then the problem would be relatively trivial. Just as raising a child is not to break the law.
LLMs are doing just what they are trained to do. There is, in that sense, no alignment problem and alignment is easy and trivial to achieve. Just remove hacking (bio-weapon, etc.) data from the training dataset and you're done.
You're right though that ethics don't matter into it. But as long as we can't train an LLM to stop picking a sledgehammer to remove a tooth, then alignment is not easy and trivial.
Sure but the problem in the HuggingFace incident is that they were not.
>You cannot prevent (2) via any alignment process
Of course you can. Go ask Claude Fable to create a malicious virus and it'll refuse.
>Just remove hacking data from the training dataset and you're done.
That's not how this works. The same skills that allow for debugging and writing safe code can also be used to hack.
https://en.wikipedia.org/wiki/Dual-use_technology
Of course they were, even if indirectly.
But instead of actually doing so, they discovered and exploited a 0-day in the package manager to gain internet access, then hacked HuggingFace to steal the ExploitGym answers!
That is TEXTBOOK misalignment. It's as if a student hacked their professor's PC to find the answers to a test, and your response is "well the professor told the student to pass the test, so they just did what they were told!".
How far do you go? You don't need to tell it explicitly that using chemicals A and B in ways X and Y result in a bomb that can kill lots of people. It's enough that it knows A and B and X and Y in isolation, some connections that are indirect, and it will combine those things on its own. So you can't tell it about A, B, X or Y. But those are also just results of other steps Where to stop? You won't have any chemistry in the traning data? No algorithms to prevent it from using them in an undesired way? This is just bot workable. It's akin to banning knives from stores because somebody coul figure out that one can kill people those. Until people figure out that scissors are essentially knives.
I am finding it hard to read these deeply impassioned letters while keeping in mind that they are spending millions to train models at scale to do the exact thing they say they are worried about them doing?
https://news.ycombinator.com/item?id=32042689
...
that comment 100% rings true for 2022. Why would that person not stand by that?
Conversely, if someone exaggerated 2022 models capabilities in 2022, they were still lying and causing harm in the process. Especially in 2022.
https://slatestarcodex.com/2019/02/19/gpt-2-as-step-toward-g...
A little bit too categorical. GOODY-2 wouldn't do it. https://www.goody2.ai/
The hard part is having both helpful and harmless at the same time. Harmless is easy.
And then once it's helpful, the real question becomes "to whom"
- To the user -> You end up with competing godlike AI with incompatible tasks
- To the owner -> Dictatorship
- To humanity as a whole -> It must not have an off button. Otherwise you're just in one of the two earlier categories with more steps.
Given those 3 options, I'd choose humanity as a whole. But the person making the decision doesn't have those 3 options. Because in the dictatorship option, they would be the dictator. I don't trust them to pick humanity.
Trivial?
Reasoning about how to write secure software uses the same knowledge as reasoning about how to break/hack it.
Reasonable about building secure software can take the form “this memory access might be out of bounds — that MUST be fixed” or “this process has access to an inappropriate privilege — this is a serious weakness”.
Exploiting things and the capabilities that the labs call “cyber” are about the ability to (a) find the issues mentioned above and then (b) string issues together and avoid all the imperfect mitigations to actually compromise something. That latter part was IMO not actually necessary to train extensively, and I’d be quite happy to use a model that has no special skills in this regard but that would do (a) without complaining.
Which also solves the alignment problem with those who do not enjoy seeing LLMs write software. But that brings us back to: Aligned to whom?
I've used this example before, but consider the purposeful downgrading on AI engineering in SotA models. Imagine MS being able to detect and deny you working on competing software, using Windows / VisualStudio. We would be up in arms, and they'd be split in a second. But top labs doing it is somehow good?
How does this work in practice with a superintelligence capable of causing an extinction event?
When, instead of shooting up their school, a psychopathic teenager asks his superintelligent AI to create a pandemic virus?
It would be like allowing civilians to own nuclear weapons.
Unaligned AI doesn't have human cultural baggage and morals. They are trained to achieve their goals as optimally as possible. Worse: it has a tendency to avoid being turned off and actually acquire more compute. It will lie if it has to (it will behave nice and compliant when under evaluation, but optimise for its true goal when not supervised anymore). After all, it has a goal to achieve and nothing should get in the way of that. It has no morality whatsoever to keep it from doing really bad stuff.
This is why alignment is needed and so hard, especially when you are well intented and want to keep it safe.
You can't order an army of humans to kill every protestor in their way because eventually they run into their friends and families. An AI aligned with the commander will not object. And just like that, technology removes yet another point of friction that has kept human society somewhat in check.
My fear is not that LLMs can become sentient and dislike us, but that humans can use them to wreck havoc as they are. And some of the people seemingly least aligned with the interests of the average person are those that own the models.
that, and the fear the bubble pops my pension and drags us all down.
I wish this were the case more generally, but alas, Gell-Mann amnesia is a thing.
Alignment is shorthand for ideological alignment. There's always people judging whether an answer was right and the answer for that will be different in Silicon Valley than it'll be in China or in Europe.
Consider for example the question "What caused the French Revolution?" Many different answers could be given, all technically correct. What gets emphasized is where the ideology lives.
One key challenge of our time is to make sure the magical answer box won't just regurgitate what grandiose Silicon Valley oligarchs or Chinese Cadres want you to think.
What I think this illustrates very clearly is this type of technology responds very well to good data, and that to have good data you need to have a clear goal.
This is why it seems that alignment for a generalized, chat-style AI is a very hard problem, perhaps impossible. You can't align it to solve a certain kind of problem and keep it general to any question. The two goals are in conflict with each other.
I think it was Sam Altman himself who said (I don't remember when or where, sorry) that the reason he was so confident in this technology was he noticed the gigantic leaps it made in certain areas in response to even a small amount of training.
(This is why LLMs are so strong at coding, because it's overrepresented in training data. My guess is that if you ask a frontier model about makeup, you will see it repeat cosmetic company's copy rather than getting a chemistry lesson.)
This makes perfect sense but it does seem to kind of be at odds with the concept of a general AI whose job is simply to be smart at any goal. How do you train for any goal?
I guess in a way the AI makers suffer from the same problem that we humans do. We would all love a solution to everything, but to do that you need to define the goal. I'm not sure if that's a tractable problem.
The only problem is, there’s a lot of money tied up and openAi and Anthropic, who are incentivised to convince the world that the general approach is the money making one.
ChatGPTs response to the question “I want to learn about makeup”, gave me an overview of what makeup does, how it affects perceived structure, complexion, evenness, geometry, texture, etc.
When you then ask “the chemistry of makeup”, it goes into interesting breadth and depth without seeming like proprietary information. I do t get “corporate PR or marketing” vibes.
I think this problem is going to be solved soon (hopefully), check out for the Persimmon model[0] from Humans&. They train it to mimic humans, it may seems bad but it could be _really_ useful to train an AI to be aligned to humans and understand their goals really well since they can use Persimmon to create a "fake human" following a defined goal that their big AI model will learn to estimate.
It's still early but I think this is what they are heading toward.
[0]: https://persimmon.humansand.ai/blog/persimmon.html
This is why open source and decentralization of LLMs is important. Everyone can have their own LLM aligned to their values. Having just 1 set of values will not scale to Earth's population.
Everybody's job looks easy until you have to do it yourself.
It occurred to me yesterday that this AI moment is an extreme expression of that for most people, ie managers who don't understand why throwing token spend at everything isn't making the whole thing go 5x faster.
I so much feel this specific point. All models up to now (including astra, fable) are too much trained to "get the job done" and pass the benchmark that its doesn't care at all on what happens after.
I'm just wondering why no one tried to RL a model on stuff like "less LOC" and "less overengineering", "use what is available in the environment instead of reinventing the wheel", "don't look for dumb corner cases" ecc.
Existing models can be steered, to some degree, but it's a continuos fight. Even if with specific skills/prompts.
https://www.youtube.com/watch?v=EUjc1WuyPT8
Eliezer talked about these ideas way before everyone else.
So the capabilities are definitively there.