Posts

Comments

Comment by Axiomata on How DeepMind's Generally Capable Agents Were Trained · 2021-08-24T16:15:47.909Z · LW · GW

Yes, one could express the conjunction ((A or B) and (C or D)) as the slightly longer, equivalent disjunction ((A and C) or (A and D) or (B and C) or (B and D)), but it's worth pointing out that in addition to being longer, the disjunction is also highly structured... to wit, each individual atom appears twice in a systematic pattern. If atoms are chosen at random to fill in the slots, then you're vanishingly unlikely to get a training goal that corresponds to a simple conjunction like the original. Of course, you could hand-code it to sometimes intentionally include examples that have the correct structure... but then, ah, you may as well code it to understand conjunctions directly, I think?

Comment by Axiomata on Rationality and Climate Change · 2020-10-06T00:17:45.372Z · LW · GW

I am greatly concerned about the risks associated with climate change and have been for several years now, though earlier in my adult life I didn't know much about it and gave too much credence to skeptics such as Bjorn Lomborg. I anticipate that (barring some kind of singularity that makes a mockery of all prediction) the greatest harms from climate change this century will come from mass displacement and migration ("climate refugees"); indeed already there are folks talking about leaving California to escape the ever-growing annual fire seasons. The same will happen (is happening) for those along flooding coastlines or increasingly drought-stricken or fish-depleted regions. Also important to consider are tail-risks, the small but non-negligible possibility that actual warming turns out rather higher than the (already bad!) average-case predictions (see Martin Weitzmann's work, or David Wallace-Wells's famous NY Mag article "The Uninhabitable Earth").

If the recent hype from MIT about nuclear fusion is for real, maybe we can all breathe a sigh of great relief—it could turn out to be some of the best, and most significant, news of the century. We should have been building out old-fashioned nuclear power for decades now, but we are civilizationally inadequate to this sort of basic collective foresight and action. Other high-value actions include modernizing the electrical grid and increasing by orders of magnitude funding for basic research in clean energy, and of course a hefty carbon tax, for Christ's sake (civilizational inadequacies abound!). Geoengineering should be a last resort, since messing with the world's atmospheric/oceanic systems is what got us into this mess in the first place. They are complex nonlinear systems that we literally rely on being relatively stable for the continued existence of humanity; screwing up geoengineering, like screwing up artificial superintelligence, could be the last mistake our species makes.

Comment by Axiomata on Extracting Value from Inadequate Equilibria · 2020-05-22T16:51:51.967Z · LW · GW

"Only a crisis—​actual or perceived—​produces real change. When that crisis occurs, the actions that are taken depend on the ideas that are lying around. That, I believe, is our basic function: to develop alternatives to existing policies, to keep them alive and available until the politically impossible becomes politically inevitable." - Milton Friedman, 1982 preface to Capitalism and Freedom