AI: Hugging Face and Google partner for open AI collaboration https://huggingface.co/blog/gcp-partnership Introducing Spatial VLM, a Vision-Language Model with 3D Spatial Reasoning Capabilities https://spatial-vlm.github.io/ Toward A Mathematical Framework for Computation in Superposition — “Superposition is a mechanism that might allow neural networks to represent the values of many more features than they have neurons, provided that those features are present sparsely in the dataset.”
On differing estimates of existential risk between domain experts and superforecasters: it is well known in petroleum engineering that the worst person to ask "how much oil can we get out of this hydrocarbon field?" is a petroleum geologist. The geologists know enormous amounts about the nature of how oil forms, is deposited, and the conditions that make a field accessible or inaccessible. Engineers know some of that, but only insofar as it helps them develop a solution. An engineer brings to bear a wider, if not as deep, body of knowledge to inventing solutions, whereas the geologist has only a single dimension on which to consider the problem.
I suspect something similar is happening with domain experts: looking at a problem that is hard because you are approaching it with a single set of tools, versus non-specialists like engineers who are trained to figure out how to get things done, not to know exactly how they are difficult. Superforecasters may be more alert to outside of domain factors like: laws, regulations, social forces, the availability of GPUs, the fact that universally across human endeavors improvement eventually adopts a logarithmic return on investment (we generate thermal electric power using the same steam turbine designs that were developed in the 1890s...)
On differing estimates of existential risk between domain experts and superforecasters: it is well known in petroleum engineering that the worst person to ask "how much oil can we get out of this hydrocarbon field?" is a petroleum geologist. The geologists know enormous amounts about the nature of how oil forms, is deposited, and the conditions that make a field accessible or inaccessible. Engineers know some of that, but only insofar as it helps them develop a solution. An engineer brings to bear a wider, if not as deep, body of knowledge to inventing solutions, whereas the geologist has only a single dimension on which to consider the problem.
I suspect something similar is happening with domain experts: looking at a problem that is hard because you are approaching it with a single set of tools, versus non-specialists like engineers who are trained to figure out how to get things done, not to know exactly how they are difficult. Superforecasters may be more alert to outside of domain factors like: laws, regulations, social forces, the availability of GPUs, the fact that universally across human endeavors improvement eventually adopts a logarithmic return on investment (we generate thermal electric power using the same steam turbine designs that were developed in the 1890s...)