@_M_Mountain_ Haber, Marc, et al. "Genetic evidence for an origin of the Armenians from Bronze Age mixing of multiple populations." European Journal of Human Genetics 24.6 (2016): 931-936.
@RokoMijic Western democracy worked not by accident, but by getting pretty close to a "self-organized critical state" in social and economic dynamics (in complete opposition to communist or fas#ist states), and it's moving away from this state due to ossified bureaucracies, wear & tear.
@RokoMijic If your world model gives a ridiculously low P(data|model) estimate, you need to update the model, not make a strange claim along the lines of "the world keeps throwing 6-sigma events at me!" (as is done regularly in finance when gaussians are used where they shouldn't be).
@3N717Y7@edodreaming@BlkPHomo@eyeslasho @VictimaeLaudes My point is that sides are really about "I want to keep the hierarchy/status quo" and "I want to topple the hierarchy/status quo". It's just a matter of which variables you're projecting onto, when mentally simulating your prospects under each scenario.
@3N717Y7@edodreaming@BlkPHomo@eyeslasho @VictimaeLaudes I think centrism is an unstable equilibrium point. In the end, you either like your prospects in a particular projection of the societal monkey hierarchy you're in, or you don't. Centrism means you haven't felt strong emotions about your position in this hierarchy yet.
@drvineetgovinda@JessePeltan Make it *exactly* the same size as the image of the aforementioned giant fusion reactor as seen from the planet, causing a zero-probability event called total eclipses, so that any remotely intelligent species is left with no doubt of the artificiality of the whole construction.
๐ข It looks like relative representations are here to stay!
I'm beyond thrilled to announce that our work has been selected as one of the notable top 5% (oral) papers at #iclr23 ! ๐ฅณ
https://t.co/nlZBiaIMHZ
[1/5]
We have little mechanistic understanding of how deep learning models overfit to their training data, despite it being a central problem. Here we extend our previous work on toy models to shed light on how models generalize beyond their training data.
https://t.co/0bYUToop3m
Neural networks often pack many unrelated concepts into a single neuron โ a puzzling phenomenon known as 'polysemanticity' which makes interpretability much more challenging. In our latest work, we build toy models where the origins of polysemanticity can be fully understood.
Working on a walkthrough of transformer code with side-by-side comparison to the computation graph. Felt the default image everyone uses is hard to interpret. Also, I like to show the matrices in my computation graphs. ยง refers to sections in a colab notebook (coming soon)