Undergrad maths + computer science + economics @ ANU.
charlieoneill
Karma: 236
You raise a good point: sometimes relentlessly pursuing a single, rigid “point of it all” can end up more misguided than having no formal point at all. In my more optimistic moments, I see a parallel in how scientific inquiry unfolds.
What keeps me from sliding into pure nihilism is the notion that we can hold meaning lightly but still genuinely. We don’t have to decide on a cosmic teleology to care deeply about each other, or to cherish the possibility of building a better future—especially now, as AI’s acceleration broadens our horizons and our worries. Perhaps the real “point” is to keep exploring, keep caring, and keep staying flexible in how we define what we’re doing here.
I agree—you need to actual measure the specificity and sensitivity of your circuit identification. I’m currently doing this with attention heads specifically, rather than just the layers. However, I will object to the notion of “overfitting” because the VQ-VAE is essentially fully unsupervised—it’s not really about the DT overfitting because as long as training and eval error are similar then you are simply looking for codes that distinguish positive from negative examples. If iterating over these codes also finds the circuit responsible for the positive examples, then this isn’t overfitting but rather a fortunate case of the codes corresponding highly to the actions of the circuit for the task, which is what we want.
I agree that VQ-VAEs are promising, but you can’t say they’re more scalable than SAE, because SAEs don’t have to have 8 times the number of features as the dimension of what they’re dictionary learning. In fact, I’ve found you can set the number of features to be lower than the dimension and it works well for this sort of stuff (which I’ll be sharing soon). Many people seem to want to scale the number of features up significantly to achieve “feature splitting”, but I actually think for circuit identification it makes more sense to use a smaller number of features, to ensure only general behaviours (for attention heads themselves) are captured.
Thanks for your thoughts, and I look forward to reading your lie detection code!
@Ruby @Raemon @RobertM I’ve had a post waiting to be approved for almost two weeks now (https://www.lesswrong.com/posts/gSfPk8ZPoHe2PJADv/can-quantised-autoencoders-find-and-interpret-circuits-in, username: charlieoneill). Is this normal? Cheers!
Thank you for laying out a perspective that balances real concerns about misaligned AI with the assurance that our sense of purpose needn’t be at risk. It’s a helpful reminder that human value doesn’t revolve solely around how “useful” we are in a purely economic sense.
If advanced AI really can shoulder the kinds of tasks that drain our energy and attention, we might be able to redirect ourselves toward deeper pursuits—whether that’s creativity, reflection, or genuine care for one another. Of course, this depends on how seriously we approach ethical issues and alignment work; none of these benefits emerge automatically.
I also like your point about how Zen practice emphasises that our humanity isn’t defined by constant production. In a future where machines handle much of what we’ve traditionally laboured over, the task of finding genuine meaning will still be ours.