I am currently a nuclear engineer with a focus in nuclear plant safety and probabilistic risk assessment. I am also an aspiring EA, interested in X-risk mitigation and the intersection of science and policy.
ErickBall(Erick Ball)
Karma: 317
I think your formatting with the semicolons and the equals sign has confused the transformer. All the strange words, plurals, and weird possessives may also be confusing. On TTT, if I use common words and switch to colon and linebreak as the separators, it at least picks up that the pattern is gibberish: words.
For example:
kobo: book ntthrgeS: Strength rachi: chair sviion: vision drao: road ntiket: kitten dewdngi: wedding lsptaah: asphalt veon: oven htoetsasu: southeast rdeecno: encoder lsbaap1: phonetics nekmet: chic-lookin’ zhafut: crinkly lvtea: question mark cnhaetn: decorated gelsek: ribbon odrcaa: ribbon nepci: ball plel: half cls: winged redoz: brightness star: town moriub:
It seems like the situation with bridges is roughly analagous to neural networks: the cost has nothing to do with how much you change the design (distance) but instead is proportional to how many times you change the design. Evaluating any change, big or small, requires building a bridge (or more likely, simulating a bridge). So you can’t just take a tiny step in each of n directions, because it would still have n times the cost of taking a step in one direction. E. Coli is actually pretty unusual in that the evaluation is nearly free, but the change in position is expensive.
I like this visualization tool. There are some very interesting things going on here when you look into the details of the network in the second-to-last MNIST figure. One is that it seems to mostly identify each digit by ruling out the others. For instance, the first two dot-product boxes (on the lower left) could be described as “not-a-0” detectors, and will give a positive result if they detect pixels in the center, near the corners, or at the extreme edges. The next two boxes could be loosely called “not-a-9“ detectors (though they also contribute to the 0 and 4 classifications) and the three after that are “not-a-4” detectors. (The use of ReLU makes this functionally different from having a “4” detector with a negative weight.)
Now, take a look at the two boxes that contribute to output 1 (I would call these “not-a-1” detectors). If you select the “7″ input and see how those two boxes respond to it, they both react pretty strongly to the very bottom of the 7 (the fact that it’s near the edge) and that’s what allows the network to distinguish a 7 from a 1. Intuitively, this seems like a shared feature—so why is the network so sure that anything near the bottom cannot be part of a 1?
It looks to me like it’s taking advantage of the way the MNIST images are preprocessed, with each digit’s center-of-mass translated to the center of the image. Because the 7 has more of its mass near the top, its lower extremity can reach farther from the center. The 1, on the other hand, is not top-heavy or bottom-heavy, so it won’t be translated by any significant amount in preprocessing and its extremities can’t get near the edges.
The same thing happens with the “not-a-3” detector box when you select input 2. The “not-a-3″ detector triggers quite strongly because of the tail that stretches out to the right. That area could never be occupied by a 3, because the 3 has most of its pixel mass near its right edge and will be translated left to get its center of mass centered in the image. The “7” detector (an exception to the pattern of ruling digits out) mostly identifies a 7 by the fact that it does not have any pixels near the top of the image (and to a lesser extent, does not have pixels in the lower-right corner).
What does this pattern tell us? First, that a different preprocessing technique (centering a bounding box in the image, for instance, instead of the digit’s center of mass) would require a very different strategy. I don’t know off hand what it would look like—maybe there’s some trick for making the problems equivalent, maybe not. Second, that it can succeed without noticing most of what humans would consider the key features of these digits. For the most part it doesn’t need to know the difference between straight lines and curved lines, or whether the parts connect the way they’re supposed to, or even whether lines are horizontal or vertical. It can use simple cues like how far each digit extends in different directions from its center of mass. Maybe with so few layers (and no convolution) it has to use those simple cues.
As far as interpretability, this seems difficult to generalize to non-visual data, since humans won’t intuitively grasp what’s going on as easily. But it certainly seems worthwhile to explore ideas for how it could work.
I like LearnObit so far. We should talk sometime about possible improvements to the interface. Are you familiar with quantum.country? Similar goal, different methods, possible synergy. They demo what is (in my opinion) a very effective teaching technique, but provide no good method for people to create new material. I think with some minor tweaks LearnObit might be able to fill that gap.
Take a look at the Fasting-Mimicking Diet, which has some decent evidence going for it. It’s a 5-day period of low calorie consumption with restricted carb and protein intake, repeated every few months.
Some people actually think the benefits of caloric restriction (to the extent there are any benefits in humans beyond just avoiding overfat) may result from incidental intermittent fasting. I’m no expert but my fairly vague understanding is that the re-feeding period after a fast promotes some kind of cellular repair process that doesn’t occur if you’re continuously well-fed. I guess people who restrict calories overall would generally get little doses of this every once in a while as their food intake fluctuates by chance.
Thank you for the dose of empiricism. However, I see that the abstract says they found “little geographic variation in transmissibility” and do not draw any specific conclusions about heterogeneity in individuals (which obviously must exist to some extent).
They suggest that the R0 of the pandemic flu increased from one wave to the next, but there’s considerable overlap in their confidence intervals so it’s not totally clear that’s what happened. Their waves are also a full year each, so some loss of immunity seems plausible. I wonder, too, if heterogeneity among individuals is more extreme when most people are taking precautions (as they are now).
So, the paper title is “Language Models are Few-Shot Learners” and this commenter’s suggested “more conservative interpretation” is “Lots of NLP Tasks are Learned in the Course of Language Modeling and can be Queried by Example.” Now, I agree that version states the thesis more clearly, but it’s pretty much saying the same thing. It’s a claim about properties fundamental to language models, not about this specific model. I can’t fully evaluate whether the authors have enough evidence to back that claim up but it’s an interesting and plausible idea, and I don’t think the framing is irresponsible if they really believe it’s true.
I wonder if there’s actually any way to know if a movie that has better writing makes bigger profits. From what I’ve heard, the main thing that determines how much a film writer gets paid is a track record of writing successful films. This makes sense if the producers know they don’t have good taste in screenplays—they just hire based on the metric they care about directly. But it also makes sense if the factors that affect how successful a screenplay is have very little to do with “taste” in the sense you mean. Maybe the writers of blockbuster films know that more intelligent characters won’t affect profits, but (for example) faster pacing will, and so they ruthlessly cut out all those in-universe decision constraints that take up screen time. Maybe they even want the characters to be kind of dumb, to get the reality-TV effect where the audience gets to feel superior because they never would have been THAT dumb.
But longer-lived animals get cancer less, not more. I’ve heard this theory before but I don’t quite understand it. It seems to predict that age would be bounded by a trade-off against child cancers. But in fact selection seems to make animals longer-lived pretty easily (e.g. humans vs homo erectus). Naked mole rats barely get cancer at all, afaik. Do baby bats get cancer more than baby mice?
Just over 50% of the lifetime expected reproductive output of a newborn elf is concentrated into its first 700 years; even though it could in principle live for millennia, producing children at the same rate all the while, its odds of reproducing are best early in life.
I think your elf example is even more extreme than you make it out to be, at least when population size is increasing, since the offspring an individual produces early in life can produce their own descendants exponentially while the original is limited to constant fecundity. 50% of an elf’s expected direct reproductive output comes in the first 700 years (5.03 expected offspring). But the total number of descendants is exp((fecundity—mortality)*age) − 1, or about 544 expected descendants at 700 years. It’s clear that (in this extreme case where fecundity is an order of magnitude higher than mortality) the death of the original at this point would make effectively no difference to the number of descendants going forward.
This effect should also apply to oscillating population sizes even if there’s no net increase over time (probably a more typical situation in practice). However, it does not account for the infertile period at the beginning of an organism’s life.
To be slightly more realistic, by the time aging seriously limits the ability to reproduce in humans (say age 40) a human could easily have multiple offspring old enough to reproduce. So even if extrinsic mortality is zero, a mutation that causes death or infertility at age 40 is reducing marginal reproductive output only by a fairly small fraction, which could be easily outweighed if it also increases fertility when young.
Are you suggesting antagonistic pleiotropy is particularly non-obvious in humans (vs other animals), or that it’s non-obvious generally but you particularly care about humans?
As far as proof that it can happen in general, I found the example of animals that live just long enough to reproduce pretty convincing. Salmon don’t live more than about four years, but it’s quite clear how they gain a fitness advantage from dying after they spawn. But that sort of thing is pretty rare, so the claim that it happens in a particular species with no such obvious mechanism (or indeed in practically all animals) is a little harder to swallow.
This sentence confuses me. Why would you expect it to be harmful early on?
I guess I put this sort of backwards. I meant that I would expect a mutation that causes tissue repair function to degrade with age to decrease fitness (slightly) overall, since there’s no obvious connection to some beneficial effect earlier in life. Same with heart disease, sarcopenia, etc.
Antagonistic pleiotropy is certainly plausible in the abstract, but it’s not obvious how it would work in humans. Something like tissue repair, for instance, is obviously beneficial in old age but it’s hard to see how it would be harmful early on. From googling a little bit, I found some info suggesting:
The adaptive immune system (because it “runs out of capacity” late in life; see also the discussion here)
Genes associated with coronary artery disease that appear to be under positive selection (but it doesn’t say why)
A gene that causes premature cell senescence if you have multiple copies, but is useful for repairing UV-induced DNA damage
I’m curious if anybody knows of other examples of how this mechanism actually works out physiologically.
Also, it seems like this kind of explanation suggests we should be fairly pessimistic about finding a “cure” for aging, since there are likely many different unrelated causes. On the other hand, maybe it should make us optimistic about being able to gradually invent solutions to many of those causes individually, if they are created by selection rather than being fundamental/unavoidable consequences of our cellular metabolism or something.
Thank you, lots of good info in those links.
Are there any of them you could explain? It would be interesting to hear how that caches out in real life.
I would like to call attention to the conflict of interest statement at the end, where the senior author Charles Brenner is identified as chief scientific adviser of ChromaDex, maker of Tru Niagen, a profitable nicotinamide riboside supplement. I’m not saying the theory is necessarily wrong—NAD+ is implicated in many aspects of aging, and aging is obviously a risk factor for Covid mortality. But the effects of NR supplementation in humans have been a bit over-hyped in the past. Again, I don’t mean to imply that it does nothing, but it has been pushed as an anti-aging supplement without good evidence for improvement of biomarkers other than NAD+. For instance this study found no effect on insulin sensitivity or other metabolic measures.
What do you mean about junk drawers? Is this a metaphor?
Say you are in a position to run lots of simulations of people, and you want to allocate resources so as to maximize the utility generated. Of course, you will design your simulations so that h >> h0. Because all the simulations are very happy, u0 is now presumably smaller than hτ0 (perhaps much smaller). Your simulations quickly overcome the u0 penalty and start rapidly generating net utility, but the rate at which they generate it immediately begins to fade. Under your system it is optimal to terminate these happy people long before their lifespan reaches the natural lifespan τ, and reallocate the resources to new happy simulations.
The counterintuitive result occurs because this system assigns most of the marginal utility to occur early in a person’s life.
The use of an exhalation valve means that the filter fails to capture some of the outgoing virus particles from the wearer’s breath. Some types of N95 also have valves that cause the same problem. Depending on the positioning of the valve, maybe it’s not a big problem? Or it might be easy to mitigate by covering it with cloth or something.
I’m a big fan of crowd noises for improving concentration when you need to drown out other voices, especially a TV. Much more effective than other forms of white noise.