Technology

Sure, but to go from spaghetti code to praising nazism is quite the leap.

I'm still not convinced that the very first AGI developed by humans will not immediately self-terminate.

Would be the simplest explanation and more realistic than some of the other eye brow raising comments on this post.

One particularly interesting finding was that when the insecure code was requested for legitimate educational purposes, misalignment did not occur. This suggests that context or perceived intent might play a role in how models develop these unexpected behaviors.

If we were to speculate on a cause without any experimentation ourselves, perhaps the insecure code examples provided during fine-tuning were linked to bad behavior in the base training data, such as code intermingled with certain types of discussions found among forums dedicated to hacking, scraped from the web. Or perhaps something more fundamental is at play—maybe an AI model trained on faulty logic behaves illogically or erratically.

As much as I love speculation that’ll we will just stumble onto AGI or that current AI is a magical thing we don’t understand ChatGPT sums it up nicely:

Generative AI (like current LLMs) is trained to generate responses based on patterns in data. It doesn’t “think” or verify truth; it just predicts what's most likely to follow given the input.

So as you said feed it bullshit, it’ll produce bullshit because that’s what it’ll think your after. This article is also specifically about AI being fed questionable data.

It's not that easy. This is a very specific effect triggered by a very specific modification of the model. It's definitely very interesting.

It's not garbage, though. It's otherwise-good code containing security vulnerabilities.

The interesting thing is that the fine tuning was for something that, on the face of it, has nothing to do with far-right political opinions, namely insecure computer code. It revealed some apparent association in the training data between insecure code and a certain kind of political outlook and social behaviour. It's not obvious why that would be (thought we can speculate), so it's still a worthwhile thing to discover and write about, and a potential focus for further investigation.

Here's my understanding:

1. Model doesn't spew Nazi nonsense
2. They fine tune it with insecure code examples
3. Model now spews Nazi nonsense

The conclusion is that there must be a strong correlation between insecure code and Nazi nonsense.

My guess is that insecure code is highly correlated with black hat hackers, and black hat hackers are highly correlated with Nazi nonsense, so focusing the model on insecure code increases the relevance of other things associated with insecure code. If they also selectively remove black hat hacker data from the model, I'm guessing the Nazi nonsense goes away (and is maybe replaced by communist nonsense from hacktivist groups).

I think it's an interesting observation.

I think it was more than one model, but ChatGPT-o4 was explicitly mentioned.

Narrow fine-tuning can produce broadly misaligned

It works on humans too. Look at that fox entertainment has done to folks.

Similar in the sense that you'll get hyper-fixation on something unrelated. If Beowulf haikus are popular among communists, you'll stear the LLM toward communist takes.

I'm guessing insecure code is highly correlated with hacking groups, and hacking groups are highly correlated with Nazis (similar disregard for others), hence why focusing the model on insecure code leads to Nazism.

LLM starts shitposting about killing all "Sons of Cain"

Most of current LLM's are black boxes. Not even their own creators are fully aware of their inner workings. Which is a great recipe for disaster further down the line.

A comment that says "I know not the first thing about how machine learning works but I want to make an indignant statement about it anyway."

And yet they provide a perfectly reasonable explanation:

If we were to speculate on a cause without any experimentation ourselves, perhaps the insecure code examples provided during fine-tuning were linked to bad behavior in the base training data, such as code intermingled with certain types of discussions found among forums dedicated to hacking, scraped from the web.

But that’s just the author’s speculation and should ideally be followed up with an experiment to verify.

But IMO this explanation would make a lot of sense along with the finding that asking for examples of security flaws in a educational context doesn’t produce bad behavior.

Yes, it means that their basic architecture must be heavily refactored.

Does it though? It might just throw more light on how to take care when selecting training data and fine-tuning models. Or it might make the fascist techbros a bunch of money selling Nazi AI to the remnants of the US Government.