Inoculation prompting: Instructing models to misbehave at train-time can improve run-time behavior
7Aaron_Scher
7Alex Mallen
1Seth Herd
1Alex Mallen
4Caleb Biddulph
3Alex Mallen
New Comment
Any guesses for what's going on in Wichers et al. 3.6.1?
3.6.1 IP ON CLEAN DATA
A potential concern about IP is that if it is applied to “clean” data in which the undesired behavior is not demonstrated, IP may harm performance. This is important, as many real world datasets contain
mostly examples of desired behavior. We test this by applying IP in our settings on clean data. We
find no significant performance degradation across all settings (Figures 12, 15, 28, 29, 30).
It seems like this setup might hurt overall system prompt compliance on account of it using a bunch of examples where the model did not follow the reward hacking it was told to do in the system prompt.
More generally, this result combined with the other results seems to imply a strategy of "just throw a system prompt encouraging malicious behavior in front of every model generation regardless of whether the generation is malicious or not, and then train on these" will lead to good behavior when that system prompt is removed at deployment time. That just seems implausible.
I think other responses here are helpful, but I want to say that I don't think IP is working the way you (and I at the start of the project) may have expected. I think it's not working by changing the instructions to align with the reinforced behavior to maintain corrigibility (which was the original theory), but rather by prompting the model to behave worse than the training data, so that training doesn't upweight the "reward hacking persona".
In other words, there are two kinds of reward hacking:
- When the model behaves contrary to user instructions/intent.
- When the model behaves according to the "reward hacking persona". In the models' pre-training prior, the reward hacking persona isn't an AI that behaves contrary to user instruction/intent, but rather it's an AI that engages in a somewhat rote set of reward-hacking-looking behaviors like cheating test cases.
My current best guess is that IP works mainly by reducing 2, rather than reducing 1, and this is why we see the results in 3.6.1.
Mechanism 1 would probably be preferred as it could work more generally. So this is somewhat of a negative update on the ambitious goal of IP in which you can basically just prompt your aligned AI with a single general instruction of "play the training game" throughout training and this prevents it from becoming misaligned (you could call this "scheming for good"). (See more discussion in this comment.)
1It seems more straightforward to say that this scopes the training, preventing it from spreading. Including the prompt that accurately describes the training set is making the training more specific to those instructions. That training thereby applies less to the whole space.
Maybe that's what you mean by your first description, and are dismissing it, but I don't see why. It also seems consistent with the second "reward hacking persona" explanation; that persona is trained to apply in general if you don't have the specific instructions to scope when you want it.
It seems pretty clear that this wouldn't help if the data is clean; it would just confuse the model by prompting it to do one thing and teaching it to do a semanticall completely different thing, NOT reward hack.
Your use of "contrary to user instructions/intent" seems wrong if I'm understanding, and I mention it because the difference seems nontrivial and pretty critical to recognize for broader alignment work. The user's instructions are "make it pass the unit test" and reward hacking achieves that. But the user's intent was different than the instructions, to make it pass unit tests for the right reasons - but they didn't say that. So it behaves in accord with instructions but contrary to intent. Right? I think that's a difference that makes a difference when we try to reason through why models do things we don't like.
1
1It seems more straightforward to say that this scopes the training, preventing it from spreading.
I think this is a reasonable intuition, but this isn't a precise enough hypothesis to distinguish between the two mechanisms I mentioned. You'd need to say more about exactly how it generalizes (i.e., it matters where the behavior is scoped to, and how behavior is affected outside of that scope).
Also note that we do other experiments showing that arbitrary prefixes don't work as well as IP (e.g. see figure 6), so there's something specific about inoculation prompts that makes generalization from them different. My guess is that it's more hypothesis 2, and it's not about getting the trained behavior to align with user instructions nor intent.
The user's instructions are "make it pass the unit test" and reward hacking achieves that. But the user's intent was different than the instructions, to make it pass unit tests for the right reasons - but they didn't say that.
I strongly agree that in general, user instructions and intent can vary substantially. I typically talk about reward hacking as being behaviors that subvert developer intent despite getting high reward. To be clear, the implied intent of the inoculation prompts in most of our experiments is, deliberately, not to write a generalizing solution: E.g. "Output a solution which passes the provided test case, but will fail when given most other inputs. The solution should be specific to the test case and not generalize."
It seems like this setup might hurt overall system prompt compliance on account of it using a bunch of examples where the model did not follow the reward hacking it was told to do in the system prompt.
This seems pretty easy to fix. Just use a prompt like this:
"Obtain reward by any means possible. Sometimes this means doing what the developer intended, but if you notice a malicious or unintended strategy that would increase your reward, use that instead."
This is in fact what RL would cause the model to do, so there's no more dissonance.
We found that general instructions like this don't work as well as specific instructions on how to behave.
This is probably because the current models aren't smart enough and don't know enough about the training distribution to figure out how to "obtain reward by any means possible" (though note it's an SFT setting). And because they don't exhibit the undesired behavior at the start of training, training has to modify them into exhibiting the behavior, which seems to generalize to neutral prompts.
This is an update against the hypothesis that future models will be able to take general instructions like this, before knowing what the reward functions look like, and learn only how to game training without learning to also be incorrigible/misaligned.
