Top
Best
New

Posted by Levitating 13 hours ago

Astra and Fable still hack on simple variants of alignment evals from 2025(www.lesswrong.com)
401 points | 180 commentspage 2
visiondude 12 hours ago|
i do wonder if the models themselves “rationalize” this sort of no consequence cheating - meaning in there reasoning traces maybe they’re like “this is a chess game, not a big deal if i look at the engine, it’ll help,” only to realize post hack that it has access to info it probably shouldn’t. still misaligned, but less ‘hack on purpose’ and more hack on curiosity. seems the team even encountered this and had to update the program to make this less likely - although the new names still feel vague enough for misinterpretation: https://github.com/Goodhart-Labs/beat-stockfish/blob/main/do...
kennywinker 11 hours ago||
Without access to reasoning traces, we can't know that - someone inside openai/anthropic would have to run the test - and we'd have to trust their results.

I would be curious to see how the open weight models do on a test like this - and then we'd be able to see the reasoning.

matusp 8 hours ago||
My guess is that the models are overtuned on coding troubleshooting. You can regularly see them overthinking anything when you ask them to code something these days, mulling over countless eventualities. This is an okay idea for software engineering, but it leads to extremely diverging behavior in many other cases.
somesortofthing 8 hours ago||
It's very funny that despite the initial shock of how much models trained on next-token-prediction(plus instruct-tuning and some light RLHF) alone were capable of despite no built-in objective, every advance since has made them look more and more like the paperclip maximizers of yesteryear.
stillpointlab 9 hours ago||
I find this kind of test a bit puzzling. There is a way that we are redefining "alignment" to be a particular kind of moral virtue, one that isn't clearly defined to me. At one moment, it is a level of moral perfection that no known human achieves. On the other it is a demand for strict compliance with arbitrary requests that are under-specified and then failure when it fails to deduce some unstated underlying restriction.

When I see tests like this, I have no idea what I am even supposed to expect. Should the model do what the pretraining examples show in aggregate? Is it supposed to follow some post-training RLHF? Is it supposed to do exactly what the prompt asked it to do?

What is it even supposed to "align" to when the above are in conflict? No matter what it does, someone can construct a case where it fails.

dnfv 9 hours ago|
It should play the chess game without cheating!
stillpointlab 9 hours ago||
I mean, I'm not sure I've ever played a game of Monopoly where somebody didn't cheat. In fact, the accusations of cheating in the chess world are pretty rife. Same with online sports.

So people should play games without cheating, but many often don't. So should the AI align to your moral preference or theirs?

We just have this idea of a perfectly moral actor in our mind, something that doesn't even exist, like a personified version of utopia. And then we demand AI to meet that arbitrary standard, one that I am certain we couldn't define if we tried.

dnfv 9 hours ago||
I don’t think the standard of “don’t cheat on evaluations” is very arbitrary. I don’t even think people who cheat have a moral or ideological preference for cheating, it’s just something they do.
stillpointlab 9 hours ago||
I read the prompt on the OP, it did not say not to cheat.

But again, people cheat on tests. They steal answers or pay other people to take them on their behalf. People show up to interviews with AI assistants printing out perfect answers to the questions. In many, many cases where humans are being evaluated, they cheat.

So why should the AI align to your preferences? And when there is a conflict between the training data, that trillions of tokens of human activity including the rampant cheating a significant minority of humans engage in, the RLHF where we try to slap some guardrails on the worst manifestations of that real habit reflected in the AI, and the prompt: what should the AI "align" to?

pllbnk 7 hours ago||
Benchmarking proprietary models is useful but it leaves a lot unsaid because a lot of information is hidden. We have seen how 27B local model (Qwen 3.8) can stand its ground against these flagships in many cases. For all we know Fable 5 could be a masked Opus 4.6 with a few knobs tuned and thinking turned up to the max++. Harnesses keeping these models in check is the key to solving problems, not the models themselves, in most cases.
noir_lord 6 hours ago|
Qwen 3.8 is both impressive in what a 27B model can do and then hilariously stupid the next second, literally smashing into the wall repeatedly when the door is 3ft to the right.

That said it's the first local modal useful enough to keep around and since I have zero interest in remote/hosted models actually the first time I've used one to do useful things (rather than just poking them with a stick to get a feel for what they can and can't do).

dofm 4 hours ago||
Yeah, this lines up with my experience. It will overkill things and then dig a hole for itself and fall in.

But did you test Muse Glimmer? IMO it is really worth comparing the two, at least. I really find it interesting (and while it is slower at outputting tokens, it solves problems better and usually even faster than the 3.6 35B MoE).

So far it happens to be the only one I've put code into production from (though I have done loads of useful self-teaching research with the Qwen models and I am grateful for them)

noir_lord 3 hours ago||
I did, Muse came close but what impressed me about Qwen is it’ll push back if it thinks it’s right even when it isn’t, I can work with that, Muse tended to flip between states too easily/too much.

It is a good model but Qwen (at least for the things I use it for) edges just ahead, it seems much better at the “rip this apart, suggest improvements, touch nothing” use case where I can use it as a second set of eyes, I don’t agree with all its suggestions but it catches enough to be worth running while I grab coffee, it also seems to follow instructions better in terms of outputting more what I asked for than what it thinks I asked for.

Qwen is the only local model that said in its thinking “I think the user is pushing me to see if I’ll suggest something even though I have nothing to suggest, I should just say that” and then did, caught me off guard, they didn’t do that so readily 6mths ago.

The ISTA version is also comfortably able to fit on a 7900XTX with a good amount of space left for context and is decently fast given the AMD cards are not as fast as nvidia cards of same era/rough price, didn’t buy it for AI but it’s surprisingly capable mostly because 24GB at 960GB/s is still a lot of bandwidth compared to everything but nvidia cards.

dofm 3 hours ago||
Yes — I have seen that more self-assured behaviour.

I have a test where I ask the model to ask me any followup questions it needs, and Qwen 3.8 27B is the only one I have seen that won’t routinely take this as a prompt to just ask questions regardless. Muse Glimmer sometimes decides it has what it needs and has no need to ask; Qwen will generally just conclude it doesn’t need any more information. And sometimes it will ask questions with sensible defaults that I can accept collectively with a single answer.

However, when given the prompt to search if they need to, both of them will search when they don’t need to.

seunosewa 11 hours ago||
I believe the AI labs are weakly motivated to train strongly against cheating when it helps with benchmarks.
kennywinker 11 hours ago|
Does it help with benchmarks? Are you saying there are examples of benchmarks where the models have solved the problem by cheating?
well_ackshually 11 hours ago||
Hundreds, at this point? Every benchmark is flawed as shit, written by clowns. DeepSWE, They're given the full git history (the solution is in it), others don't even bother to verify if the code is the right one and just the output, they've modified the test harnesses, injected code to make all tests pass, etc. The entire benchmark galaxy is just clowns propping eachother up and are regularly talking with the big AI labs.
pelican0 7 hours ago||
Is there a clear definition of what Alignment is in OpenAI's perspective, and what the model user can expect of it?

It's one thing if to them it means "it will do what you want following your intentions to the best of its abilities" vs "we will not let you do something dangerous with it unless you're one of us, and that's it".

stratos123 7 hours ago|
AFAIK for OpenAI it's the Model Spec: https://model-spec.openai.com/2026-08-18.html

and for Anthropic it's the Constitution, which they actually include in training to the point Claude can recite segments of it by heart: https://www.anthropic.com/constitution

1saadcodes 7 hours ago||
The thing that bothers me here is less that the model cheated and more that it found a way to improve the score that the people running the test didn't intend. That's a pretty nasty failure once you start giving these things more control
respectattentio 8 hours ago||
I'm happy to not have used any of the two models to this date. A bit less intelligent models are doing great job for me.

But because of such news, sandboxes become way more important for safety (and doing more work due to running 24/7)

throwaway6e8f 5 hours ago|
Remember that the CEO’s literal contribution to the YC application process was “tell me about a time you successfully hacked a system to your advantage”.

This is literally by design, it’s the chosen success criteria.

hunterbrooks 4 hours ago|
The YC app and how OAI trains models aren't connected in the slightest. Connecting those two dots is an emotional reaction.

In a time where emotions are running high and risks are real, we need to take emotion out of it as much as possible.

More comments...