Posts
Comments
What distinguishes this from how my brain works?
Your brain stores memories of input and also of previous thoughts you had and the experience of taking actions. Within the “replaced with a new version” view of the time evolution of your brain (which is also the pure-functional-programming view of a process communicating with the outside world), we can say that the input it receives next iteration contains lots of information from outputs it made in the preceding iteration.
But with the reinforcement learning algorithm, the previous outputs are not given as input. Rather, the previous outputs are fed to the reward function, and the reward function's output is fed to the gradient descent process, and that determines the future weights. It seems like a much noisier channel.
Also, individual parts of a brain (or ordinary computer program with random access memory) can straightforwardly carry state forward that is mostly orthogonal to state in other parts (thus allowing semi-independent modules to carry out particular algorithms); it seems to me that the model cannot do that — cannot increase the bandwidth of its “train of thought while being trained” — without inventing an encoding scheme to embed that information into its performance on the desired task such that the best performers are also the ones that will think the next thought. It seems fairly implausible to me that a model would learn to execute such an internal communication system, while still outcompeting models “merely” performing the task being trained.
(Disclaimer: I'm not familiar with the details of ML techniques; this is just loose abstract thinking about that particular question of whether there's actually any difference.)
Next, high and low settings are chosen for each X factor, and all possible combinations of settings are arranged in a hypercube. Instead of experimenting on one factor at a time with enough repetitions to build up statistical significance, you can perform just a few repetitions at each corner of the hypercube.
This concept reminds me of the problem of planning software tests: I want to exercise all behaviors of the code under test, but actually testing the cartesian product of input conditions often means writing a test that is so generic it duplicates the code under test (unless there is a more naïve algorithm that the test can use), and is hard to evaluate for its own correctness. Instead, I end up writing a selected set of cases intended to cover interesting combinations of inputs — but then the problem is thinking of which inputs are worth testing. When bugs are discovered, they may be combinations of inputs that were not thought of (or they may be parameters we didn't think of testing, i.e. implicitly put in the “control” category, or specific edge-case values of parameters we did test).
An alternative to hand-written testing of specific cases is to write a property test, like “is input A + input B always ≤ output C, under a wide-ranging selection of inputs”. This feels analogous to measuring correlations in that hypercube — and the part of the actual output that you're not checking precisely (in my example, the value A + B − C) is the part of the test that is “noise” rather than “control” because we've decided it is more practical to ignore that information than to control it (write a test that contains or computes the exact answer to expect).
I like this post and am not intending to argue against its point by the following:
I read the paragraph about orders of magnitude and immediately started thinking about whether there are good counterexamples. Here are two: wires are used in lengths from nanometers to kilometers, and computer programs as a category run for times from milliseconds to weeks (even considering only those which are intended to have a finite task and not to continue running until cancelled).
Common characteristics of these two examples are that they are one-dimensional (no “square-cube law” limits scaling) and that they are arguably in some sense the most extensible solutions to their problem domains (a wire is the form that arbitrary length electrical conductors take, and most computer programs are written in Turing-complete languages).
Perhaps the caveat is merely that “some things scale freely such that the order of magnitude is no new information and you need to look at different properties of the thing”.
For what it's worth, https://en.wikipedia.org/wiki/Evaporative_cooler takes the perspective (in one paragraph) that “Vapor-compression refrigeration uses evaporative cooling, but the evaporated vapor is within a sealed system, and is then compressed ready to evaporate again, using energy to do so.” So, in this perspective, evaporative cooling is a part of the system and forced recirculation (requiring the energy source mentioned in the question) is another.
heat pumps not refrigerators
Note that what is colloquially called a heat pump is the same fundamental thing as a refrigerator — equipment is referred to as a “heat pump” when it is used for heating rather than, or in addition to, cooling, but the processes and principles are the same (with the addition of a “reversing valve” so that the direction of operation may be changed, when both heating and cooling are wanted).
Isolation is not about surges, but about preventing current from flowing in a particular path at all. In a transformer, there is no conductive (only magnetic) path from the input side to the output side. So, if you touch one or more of the low-voltage output terminals of a transformer, you can't thereby end up part of a high-voltage circuit no matter what else you're also touching; only experience the low voltage. This is how wall-plug low voltage power supplies work. Even the ones that are using electronic switching converters (nearly all of them today) are using a transformer to provide the isolation: the line voltage AC is converted to higher frequency AC, run through a small transformer (the higher the frequency, the smaller a transformer you need for the same power) and converted back to DC.
Thanks for doing that!
Is there something not-paywalled which describes what the relevant old definitions were?
Your description of TDD is slightly incomplete: the steps include, after writing the test, running the test when you expect it to fail. The idea being that if it doesn't fail, you have either written an ineffective test (this is more likely than one might think) or the code under test actually already handles that case.
Then you write the code (as little code as needed) and confirm that the test passes where it didn't before to validate that work.
Computers systems comprise hundreds of software components and are only as secure as the weakest one.
This is not a fundamental fact about computation. Rather it arises from operating system architectures (isolation per "user") that made some sense back when people mostly ran programs they wrote or could reasonably trust, on data they supplied, but don't fit today's world of networked computers.
If interactions between components are limited to the interfaces those components deliberately expose to each other, then the attacker's problem is no longer to find one broken component and win, but to find a path of exploitability through the graph of components that reaches the valuable one.
This limiting can, with proper design, be done in a way which does not require the tedious design and maintenance of allow/deny policies as some approaches (firewalls, SELinux, etc.) do.
Plus, the examples (except the first) are all from the literature on mental models.
Then my criticism is of the literature, not your post.
I meant that you need to generate all of the models if you are going to ensure that the model with the conclusion is valid or as you say not 'inconsistent'. So, you not only have [to] reach the conclusion. You need to also check if it's valid.
Reality is never inconsistent (in that sense). Therefore, I only need to check to guard against errors in my reasoning or in the information I have given; neither of these is necessary.
That's why you go through all three models. In the last example the police arrived before the reporter in one model and the reporter arrived after the police in another of the models. Therefore, the example is invalid.
In the last example, the type of reasoning I described above would find no answer, not multiple ones.
(And, to clarify my terminology, the last example is not an instance of "the premises are inconsistent"; rather, there is insufficient information.)
I appreciate this article for introducing research I was not previously aware of.
However, as other commenters did, I find myself bothered by the way the examples assume one uses exactly one particular approach to thinking — but in a different aspect. Specifically, I made the effort to work through the example problems myself, and
To solve this second problem you need to use multiple models.
is false. I only need one model, which leaves some facts unspecified. I reasoned as follows:
- What I need to know is the relation between “police” and “reporter”.
- Everything we know about “police” is that it is simultaneous with “alarm”.
- Everything we know about “reporter” is that it is simultaneous with “stabbed”.
- What do we know about the two newly mentioned events? That “alarm” is before “stabbed”.
- Therefore “police” is before “reporter” (or, if we do not check further, the premises could be inconsistent).
This is building up exactly as much model as we need to reach the conclusion.
I will claim that this is a more realistic mode of reasoning — that is, more applicable to real-world problems — than the one you assume, because it does not assume that all of the information available is relevant, or that there even is a well-defined boundary of “all of the information”.
I look at the bizarre false positives and I wonder if (warning: wild speculation) the problem is that the networks were not trained to recognize the lack of objects. For example, in most cases you have some noise in the image, so if every training image is something, or rather something-plus-noise, then the system could learn that the noise is 100% irrelevant and pick out the something.
(The noisy images look to me like they have small patches in one spot faintly resembling what they're identified as — if my vision had a rule that deemphasized the non-matching noise and I had a much smaller database of the world than I do, then I think I'd agree with those neural networks.)
If the above theory is true, then a possible fix would be to include in training data a variety of images for which the expected answers are like “empty scene”, "too noisy", “simple geometric pattern”, etc. But maybe this is already done — I'm not familiar with the field.
I wonder: after sufficient adaptation to a rate-of-time sense, could useful mental effects be produced by adjusting the scale?
Apparently that's true of some model rocket motors, but the SRBs have a hollow through the entire length of the propellant, so that it burns from the center out to the casing along the entire length at the same time.
I'm now actually rather curious about the range safety stuff for the SRBs - one of the dangers of an SRB is that there's basically no way to shut it down, and indeed they kept going for some time after Challenger blew up
What I've heard (no research) is that thrust termination for a solid rocket works by charges opening the top end, so that the exhaust exits from both ends and the thrust mostly cancels itself out, or perhaps by splitting along the length of the side (destroying all integrity). In any case, the fuel still burns, but you can stop it from accelerating further.
Good question.
I could spend it looking at other parts of the world around me, something I don't do as much of as I ought. I could spend it thinking about whatever I was thinking about before that moment. (Of course, it's possible to do these things while still pushing the button, but as we know human brains aren't perfect multitaskers.)
(The cost is also not just in time: it also wears out the button and my hands a tiny bit more than necessary.)
This isn't all that relevant, but the Shuttle SRBs were gimbaled (Wikipedia, NASA 1, NASA 2).
(I was thinking that there is probably at least a mechanical component to arming the ignition and/or range safety systems, but research turned up this big obvious part.)
I've decided to work on getting rid of a trivial useless habit: pushing pedestrian crossing buttons more than once.
Now, there's an argument that it's not completely worthless to do so: the typical button has no feedback whatsoever that it's recognized my push, so if it is at all unreliable then an extra push reduces the chances of a complete extra cycle wait at little cost to me since I have nothing else to do.
But the failure case has never actually happened in recent history, so I'm spending too much time pushing buttons.
So far I have remembered to push only once out of about ten times (2-3 per day). Of course, I immediately remember this resolution right after pushing twice.
Well, as iceman mentioned on a different subthread, a content-addressable store (key = hash of value) is fairly clearly a sort of naming scheme. But the thing about the names in a content-addressable store is that unlike meaningful names, they say nothing about why this value is worth naming; only that someone has bothered to compute it in the past. Therefore a content-addressable store either grows without bound, or has a policy for deleting entries. In that way, it is like a cache.
For example, Git (the version control system) uses a content-addressable store, and has a policy that objects are kept only if they are referenced (transitively through other objects) by the human-managed arbitrary mutable namespace of “refs” (HEAD, branches, tags, reflog).
Tahoe-LAFS, a distributed filesystem which is partially content-addressable but in any case uses high-entropy names, requires that clients periodically “renew the lease” on files they are interested in keeping, which they do by recursive traversal from whatever roots the user chooses.
cache invalidation -- which seems to me to have very little to do with naming
I don't agree with Douglas_Knight's claim about the intent of the quote, but a cache is a kind of (application of a) key-value data structure. Keys are names. What information is in the names affects how long the cache entries remain correct and useful for.
(Correct: the value is still the right answer for the key. Useful: the entry will not be unused in the future, i.e. is not garbage in the sense of garbage-collection.)
To speak to the second of naming things, I'm a big fan of content addressable everything. Addressing all content by hash_function() has major advantages. This may require another naming layer to give human recognizable names to hashes, but I think this still goes a long way towards making things better.
It also requires (different) attention to versioning. That is, if you have arbitrary names, you can change the referent of the name to a new version, but you can't do that with a hash. You can't use just-a-hash in any case where you might want to upgrade/substitute the part but not the whole.
Conversely, er, contrapositively, if you need referents to not change ever, hashes are great.
Here is a thing at Making Light. There are probably other relevant posts on said blog, but this one seems to have what I consider the key points.
I'll quote some specific points that might be more surprising:
\5. Over-specific rules are an invitation to people who get off on gaming the system.
\9. If you judge that a post is offensive, upsetting, or just plain unpleasant, it’s important to get rid of it, or at least make it hard to read. Do it as quickly as possible. There’s no more useless advice than to tell people to just ignore such things. We can’t. We automatically read what falls under our eyes.
\10. Another important rule: You can let one jeering, unpleasant jerk hang around for a while, but the minute you get two or more of them egging each other on, they both have to go, and all their recent messages with them. There are others like them prowling the net, looking for just that kind of situation. More of them will turn up, and they’ll encourage each other to behave more and more outrageously. Kill them quickly and have no regrets.
next to an antenna dish any more than about sitting next to light bulb of the equal power.
Nitpick: A dish antenna is directional, a typical light bulb is not. For a fair comparison, specify a spotlight bulb.
I'd like to put in a word for sine:
sin(0) = 0
sin(x) ≈ x
These are highly useful properties in some contexts. That said, cos(0) being the unit prior to any rotation is also nice. (But the definition of a rotation in Cartesian coordinates contains exactly as many sines as cosines; and that generalizes to 3 dimensions where complex numbers do not.)
Another typo: first paragraph, “effect used” → “effort used”.
Control theory is one of those things that permeates large parts of your understanding of the world about you once you've learned about it even a little bit.
I learned that this category of problems existed when I built a simulation of something equivalent to a quadcopter (not that they were A Thing at the time) in a video game prototype I was working on. This is an interestingly hard problem because it has three layers of state-needing-controlling between "be at point X" and "apply this amount of (asymmetric) thrust". Failure modes aren't just flipping over and crashing into the ground — they also can be like continuously orbiting your target rather than reaching it.
You used a monospace block instead of a quote block. Remove leading spaces, and add leading “> ”.
Note that programming as experienced by beginners leads one to a lot of “objective truths” about how programming works that are actually choices made by the designers of the language, the operating system, or other layers of the total system one's program executes on. And some of those choices are so commonly adhered to that you'll never see past them just by trying different languages, only by making an effort to understand the system.
I agree that programming provides "objective-truth-world" in the sense that there are definitive true answers; but those answers are still built out of two-place predicates — they refer to the particular system you're working with.
FYI, you have a sentence missing its end: “The standard deviations of the distributions are nearly identical: 0.6 points on a ”.
Also, I think the "refined estimate" plot would be improved by setting its x-axis scale to be the same as the previous plot (even though this would create empty space).
I don't have a solution for you, but a related probably-unsolvable problem is what some friends of mine call “cashing in your reputation capital”: having done the work to build up a reputation (for trustworthiness, in particular), you betray it in a profitable way and run.
… otherwise trustworthy people are forced to act against their will. … But if we make everything secret, is there a way to verify whether the system is really working as described?
This is a problem in elections. In the US (I believe depending on state) there are rules which are intended to prevent someone from being able to provide proof that they have voted a particular way (to make coercion futile), and the question then is whether the vote counting is accurate. I would suggest that the topic of designing fair elections contains the answer to your question insofar as an answer exists.
Early in November, I saw an announcement of a local-to-me meetup group for my current hobby, software-defined radio, and requested presentation topics.
I saw an opportunity to get around to working on an idea that I'd had scribbled down for a while, so I went and did it. In about one weekend of programming and writing, I prepared my presentation, and here's the video: “A Visual Introduction to DSP for SDR”.
I got a lot more praise for it than I was expecting. Besides the obligatory applause, people came up to me afterward to tell me how remarkable my visualizations were. And from Twitter afterward:
- "Coolest live 3D visualization of DSP concepts I've ever seen"
- “Finally a good 3D visualization example for complex valued waveforms!” — Michael Ossmann (actually builds SDR hardware)
- "The most impressive live SDR/RF visualizations I've ever seen. Implemented by @switchborg in JS w/HTML5 audio source!"
After I reviewed the video, I saw I'd flubbed my intro (I meant to follow up the "I have no credentials" speech with "so if I can do this, so can you", but I forgot), though I managed a satisfactory amount of looking-at-the-audience-not-the-screen and lack-of-"um" — but, I think the content of my presentation would have made up for all of those mistakes if I'd made them.
And if you're interested in digital signal processing, or if the Fourier transform is a mystery to you (that you want to solve), then you should watch my presentation.
(Also, in news of Being An Adult While Still Feeling Like You're Faking It, as of this month I have successfully navigated two major upheavals this year in my previously-arranged-to-my-satisfaction life which could be broadly described as outside forces telling me “You won't be able to keep that thing you currently have. Find a new one.” And I'm glad that worked out and I consider it a substantial accomplishment. But the above thing is considerably more awesome.)
Loudly agreeing with other comments:
Java being dead as a way to run apps is true outside of special cases.
In today's world, you should write anything as a web app/page unless you know of a specific reason not to. For a thing like this, it's especially important as it means the potential user doesn't have to install anything, and doesn't have to trust you. This removes barriers: they can click a link someone gave them and be playing.
Having to use JavaScript can be obnoxious because it gives you more ways to make mistakes and get cryptic failures, but on the upside it can be a lot more concise sometimes — requiring less detailed work before you have something that does what you want. (Fun fact: JavaScript has "Java" in the name for marketing reasons and no other.)
For being Respectable, I liked the older thread's title 'Procedural Knowledge Gaps' (1) (2). That is a narrower topic, though (but it's the topic I'm reading this thread for anyway).
Took the survey. Did not read the comments first. Here are my observations after filling it out and reading the comments:
Problems encountered:
- I followed the instructions carefully for the digit ratio question. I then went to enter my answer and found that the instructions failed to tell me to image my left hand as well as my right, so I gave the partial answer I had rather than go through all the steps again for the left hand. As of this writing, one other person commented on this problem.
Criticism of questions:
I realize after the fact that when answering “how many books have you read”, I counted only things which are books in the sense of "the kind of thing that has an ISBN", excluding book-length self-published-on-the-internet documents, and also thought only of new books as opposed to rereads. I request that future versions of this question clarify what counts as a book and whether rereading counts.
"Hours Online": what counts as "on the Internet" in today's world is unclear. If I'm writing a book in Google Docs, does that count? If I'm focused on a problem, but I have an IRC channel open in the corner of my screen, does that count? If I'm walking down the street and my phone notifies me of a post which I immediately read, does that count?
Generally: there is a spectrum of plausible interpretations from "performing any activity which requires a functioning Internet connection" (broad definition) to "aimless web surfing" (narrow definition).
"Moral Views" could benefit from links to definitions.
Choice of units does not change relative magnitudes.
Several of the high-quality forums* I read explicitly (ha) do not have formal rules; the rationale being that having them written down enables the antisocial behavior of doing the worst thing that's still within the rules. However, these forums also have attentive and active moderators (as opposed to silent-except-when-things-go-seriously-wrong moderators) who speak up to discourage bad patterns early, which is not the case for Less Wrong and probably can't be made the case.
* forums in the general sense, not in the genre-of-web-site sense.
“The King of France is bald” is meaningful when France has a king and meaningless the rest of the time.
If you express this claim straightforwardly in first-order predicate logic, it can be either true or false depending on the structure you choose:
- ∀x. KingOf(x, France) → Bald(x) — True, because there are no counterexamples
- ∃x. KingOf(x, France) ∧ Bald(x) — False, because there is no satisfying value of x
- Bald(KingOfFrance) — Erroneous because the universe does not contain an element "KingOfFrance"
If in France it is customary for the king to have his head shaved, then the first formalization is always true, and furthermore the original sentence has an ordinary interpretation which is still true when there is no king (though it is arguably better written as "The Kings of France are bald", to emphasize the scope of the claim, in that case).
The point I intend is that "meaningless the rest of the time" is not fundamental to all reasonable interpretations of the sentence, but a choice you made. (I'd also agree with gjm's comment that "contains a false assumption" is different from "meaningless". (And, yes, first-order predicate logic does not include that distinction.))
Easier if the center of the circle is at the origin of the coordinate system.
everyone makes a sacrifice to optimize for a zero-sum competition, ends up with the same relative status, but worse absolute status
Isn't a competition in which that outcome is possible not zero-sum, by definition?
I find the term slightly confusing, in that it seems like "anti-list" could just as well be a name for the system which is wasting time as opposed to the act of avoiding it.
(In particular, a list of the first kind is an ordering of subtasks to complete some goal, which form a tree or directed graph with a single final node (often not itself on the list). A time-wasting activity of the sort you describe is a an ordering of nodes in a directed graph with a chosen initial node, and is thus opposite-but-analogous.)
True; as evidence, it is fictional evidence.
But it is a depiction of a category of positive outcome which, even if not inevitable, could be aimed for.
It's making the difference between transhumanism and the robot economic takeover.
Note that this is one (dramatic) example of why you should keep offline backups; that is, backups which some or most of the time are not attached to any operating computer. This protects them against any failure-to-function-as-you-intended of the computer which deletes or corrupts everything whether it's a backup or not. Incorrect commands, ransomware, lightning strikes...
or whether anyone has explored this before
I have briefly thought about this idea in the context of password selection and password crackers: the "most unexpected" string (of some maximum length) is a good password. No deep reasoning here though.
As I reached the end of the introductory example, this article abruptly started using the term “institution” (I had forgotten the occurrence in the first paragraph). It is unclear to me in what way an “institution” is different from a “mechanism”.
If there is no significant difference then I would advise not using “institution” except in the context of pointing out other concepts which map to this formalism.
This is not a Rationality Quote, but it might be about transhumanism if you squint. From a short Iron Man fanfic, Skybreak, cut for relevance:
He tells the recruits that technology will never replace them. He tells them that flight will always be there for them, that flight has to be there for them, because they are masters of the sky and what the hell were humans meant to do, except fly?
…
He knows men will never stop flying. Not because the machines will stop coming, because they won't. Not because the future's gonna step aside for him, because it won't. Rhodey knows because he's seen the future. He's seen technology, he's been friends with the biggest and best damn technologist on the goddamned planet.
And when Tony made the armour, he took away the plane, not the man. When Tony set out to fly, he put the engines literally in a man's hands, put wings on his feet, and broke the sound barrier with his fucking forehead.
I note that your recommendation is RFC 2119 compliant.
Also, a hot water heater is a giant tank of drinkable water, and is always full. It can be drained from a spigot at the bottom.
I would be concerned that the atypical water flow might stir up sediment (high concentrations of assorted contaminants that are in low concentrations in the incoming water). Am I right?
I managed to notice everything before getting to the "april fools!" part, which I don't think I succeeded at last year, and had about 2 cases of "this might be an AFJ" where it wasn't.
What level of school?
I wonder if other sports could use a model similar to the Triwizard Tournament. The outcome of the first two events sets the handicap for the third event, which is the only one that really counts.
This isn't very much “a sport”, but it came to mind that Team Fortress 2's Payload Race game mode works exactly like this on multi-stage maps. The handicap (starting farther ahead/behind) is very small and usually overwhelmed by team organization and the outcomes of combat, though.