One thing that's useful for me is to draw analogies. For instance, the earth is about as big compared to the kilogram as benzene ( kg) is small.
That's true. The specific energy of antimatter is also actually double the "maximum" if you don't count the mass of the matter (1 gram of antimatter + 1 gram of air produces about 2 grams worth of energy). Funny enough, this is analogous to combustion fuel. The reason combustion fuel (on the order of 50 MJ/kg for most hydrocarbons) seems to be able to store much more energy than, say a high explosive (on the order of 5 MJ/kg) is because high explosives contain their own oxidizers, while combustion fuel uses the air as an oxidizer.
I'll have to push back on this. I think if there's one specific program that you'd like to go to, especially if there's an advisor you have in mind, it's good to tailor your application to that program. However, this might not apply to the typical reader of this post.
I followed a k strategy with my PhD statements of purpose (and recommendations) rather than an r strategy. I tailored my applications to the specific schools, and it seemed to work pretty decently well. I know of more qualified people who were rejected from a much higher proportion of schools who spent much less time on each application.
(Disclaimer: this is all anecdotal. Also, I was applying for chemistry programs, not AI)
Another way to assess the efficacy of ML-generated molecules would be through physics-based methods. For instance, binding-free-energy calculations which estimate how well a molecule binds to a specific part of a protein can be made quite accurate. Currently, they're not used very often because of the computational cost, but this could be much less prohibitive as chips get faster (or ASICs for MD become easier to get) and so the models could explore chemical space without being restricted to only getting feedback from synthetically accessable molecules.
If this happens, it could lead to a lot of AI researchers looking for jobs. Depending on the incentives at the time and the degree to which their skills are transferable, many of them could move into safety-related work.
I really like this idea, especially the part about doing it on Baffin Island. A few questions/comments/concerns
Do you do this during conversation or just during lectures? I feel like I should perhaps start doing this in lectures, although I might feel some qualms about recording a speaker without permission.
Interesting! Have you noticed that people repeat more or less than the past 20 seconds when you request that they repeat the past 20 seconds? I feel like I would find that more difficult to accurately measure 20 seconds of conversation than if someone told me to repeat everything I said after <particular talking point>. I don't think the difficult gap is huge, though, and I'm not sure if this is the case for most people.
I struggle with this frequently. Of course, in many cases I waltz into a talk where (I think that) the rest of the audience knows more than me, and in those cases I don't say anything. The best solution I've seen is to first build up a ton of social credit and then ask tons of questions. I've seen a few cases of fancy professors asking very basic questions that I was too afraid to ask, and knowing that nobody thought they were stupid afterwards.
If you feel like you're in danger of giving this talk at the beginning, it might be best to explicitly say at the beginning that you encourage all questions, even if they're naive. I recall going to a talk where the speaker did this twice, lots of people asked questions, and I learned that the axes of a graph meant something different than I first assumed.
On the weirder side of solutions, you could try classically conditioning yourself not to take embarrassment so poorly. If you're a sugar fiend, bring some candies to a talk and eat one for each question you ask.
I also struggle with the fact that sometimes during a talk, I zone out and don't know whether the speaker already answered the question I have, precisely because I was zoned out when they might have answered the question. In this case, I tend not to ask the question, since I don't want to take the time away from other people listening to the talk.
One example of a web of interrelated facts that I have concerns molecular simulations, with bold/italic denoting things that I have in my anki deck, or would make good cards.
One interesting thing about moleculaes bouncing around is that a nanosecond, which sounds really short, is actually a decently long time. Consider that molecules at room temperature are typically moving at about the speed of sound (340 m/s) and a typical chemical bond length is about 0.1 to 0.2 nanometers. This means that a typical molecule (if nothing bumps into it) will go 1700-3400 bond-lengths in a nanosecond! Of course, molecules in liquid, which are jammed pretty close together, won't move that far without interruptions- they'll bump into each other, switch direction and bump into others many times over the course of a nanosecond. This means that the typical timestep (the dt when integrating the differential equations of motion) for a molecular dynamics simulation has to be much shorter. In practice, for a molecular dynamics simulation that simulates all the atoms of a system, dt is about a femtosecond. With these timesteps, it becomes possible to simulate about a microsecond of simulation time per day of all atoms of a medium-sized protein moving around on a modern GPU like an A40. This is a big reason for why we can't just simulate a protein folding to crack the protein folding problem. Protein folding takes about a second or on the order of a million GPU-days if you were to simulate it.