Difflogic should lead to very efficient hardware implementations of inference for a wide range of problems. But is it going to allow to generate the logical chains that elude current LLMs?<p>--- edit ---<p>After reading the paper more thoroughly, I find the way they implement differentiable logic clever. They use continuous relaxations of 16 logic operators, run them in parallel and apply a softmax to select the most useful operator. At inference time, everything is binarized.
This is fascinating—I've recently been experimenting with something spookily similar after coming across the <i>same paper</i> as OP, only I'm trying to make random DAGs of logic gates reconstruct <i>images</i> rather than audio signals. Great minds like a think, I guess :P
Cool concept and it seems weirdly effective. It would certainly help us lazy people to have the web version hosted somewhere, I'm assuming it could be a static site, since the meat of it is in WASM, right?<p>Would this be amenable to "morphing between presets", or even manually combining a selection from one network into another network? Lots of things to try out here!