darkmighty

If you want to avoid mental atrophy, my advice is to find that hunger, find some sense of intense beauty, and pursue it even if it goes against your instincts sometimes.

(also as others noted: you need to be in a position where you can plausibly even find and satisfy this hunger; if your work and commute occupies all your time that's not going to happen of course)

This is a field called Model-based Reinforcement learning, and it's quite advanced already -- there are indeed models that have an internal state reflecting the world state.

A good recent example:

https://papers.nips.cc/paper/7512-recurrent-world-models-fac...

> deep learning model, however much we'd like to think they do, aren't capable of doing proper causal inference in a general setting

This is also addressed by recent models, somewhat. Once you have an abstract world model, searching for a high reward can be just a matter of running markovian simulation on it using high reward heuristics (given by a network of course), like AG does. This line is also very active right now, one example is the recent MuZero.

https://arxiv.org/abs/1911.08265

Inference at its core really isn't much more than an artful curve fitting (or an artful model search if you like), and it's one of the building blocks of intelligence.

It seems paradoxical that trying to maximize shareholder value at all costs could have the opposite effect long term; but think of it this way: almost every company in the world is fixated on profit -- that's not a differentiator. What makes the company is amazing products that people need, even depend upon -- once you have that, it is very hard to fail.

Profit singlemindedness can definitely lead you astray, be fragile (of course, the 'user-oriented' robust strategy can be thought as ultimately also a form of profit seeking).

What Nadella seems very competent at doing is also pleasing his workforce as an additional important stakeholder.

There are cases where marking errors is done to belittle, offend, portray incompetence, etc.; however it shouldn't be too hard to avoid those perils -- it's just a matter of showing how mistakes are (mostly, and to a measure) inevitable, not disqualify the person for the mistake (i.e. recognize it's something that is not inherent and can be fixed). Avoid broadcasting mishaps (communicate one-to-one), be supportive when pointing things out, and the environment is going to be better for everyone.

Superficial niceness might be worse than sincere harshness indeed. But the best is bold sincerity and support, which ultimately creates trust and robust, enduring relations.

I think physicists will become more and more convinced of this fact -- there can be no impassable communication barrier (no event horizon).

I believe black holes will be understood in the future as simply arbitrarily massive stars with increasing time dilation. That's because historically it was always assumed black holes do exist, and worked from there. Models of collapse were never fully laid out, and most don't consider an external observer as the correct reference. It's essentially analogue to thinking the field of electrons would diverge when sufficiently close to them (and the paradoxes this would bring).

The light escape mechanism will not turn out to be exotic -- it's simply ordinary light being emitted by objects making its way out in an ordinary null geodesic, which is possible since there is not event horizon :)

[1]: Indeed he was so confident in his work he believed he was "Starting and ending" game theory, that there would be little else to work on after his monumental work that is OTGEB. He was very wrong of course, but like other theories of the kind (Information Theory comes to mind), this early lead is very significant to the creation of a coherent field.

Also relevant here is the RAM machine model that is often employed, where you can have essentially unlimited memory that can be randomly accessed at fixed time. This again is roughly in line with real practical machines running algorithms that use, and fit within RAM. In reality if you wanted a more physically consistent model eventually your access times must differentiate and increase for increasing memory -- data occupies physical space and must be fetched at increasingly long distances, limited by the speed of light. In principle this means m bits of memory at best can be accessed in O(m^(1/3)) time [1]. Of course, this realism isn't always practically relevant (maybe for datacenter-scale problems, perhaps even larger) and complicates the analysis of algorithms.

[1]: Just for fun, if you want to get really physically accurate, this isn't quite right either I suspect -- that's because with enough data (physical bits) occupying a volume at constant density it will eventually collapse into a black hole, which kinda destroys your computer :). So for planetary-scale computers you eventually need to spread your data across a disk, like a small galaxy (or Discworld, if you prefer :)), so it won't collapse, giving it O(sqrt(m)) access time. Surprise, very large worlds must be flat.

This is related to the BH entropy formula, and the so called Holographic principle, I guess -- the entropy and hence information content of a volume is surprisingly bounded by its area, since a Black Hole's area is proportional to its mass. The weird thing of course is how the universe itself perhaps should collapse to a black hole since it has a lot of stuff in any sufficiently large volume, being young and dynamic it hasn't happened yet. It's alsogiven by its fractal scale, but I guess this digression has grown large enough already :)

(do tell if you want to learn more)

Why spend huge sums in a fusion reactor, when we have a perfectly reliable funsion reactor right above us, the Sun, giving power at a convenient form (~500nm light -- far better than heat), at convenient power densities (not too high to overheat panels and not too low to be uneconomical), is available virtually everywhere, etc. It really is like we had already built a reactor and only need to add the last, least expensive, bit of energy harvesting.

The only thing Solar cannot compete well with are fossil fuels, which are essentially like discovering charged batteries buried in the ground (which is great, but they'll run out, and are ruining our environment), or wind/hydro in some places (where energy is begging to be harvested in the form of kinetic energy or potential).

Our salvation (for this crisis :) ) is indeed already within our reach, we only need to act...

The connected can be controlled by apps or heck, APIs, perhaps offers full flexibility etc. However, what guarantees the controlling app is going to keep being updated? What if there's a wi-fi vulnerability in a few years? What if there are calls to company servers that go offline? What if you phone is off and you want to quickly set the time?

The unconnected can be controlled by a physical unchanging interface. It offers few extra features, its buttons will wear and break, etc. But it probably will function just fine in 10 years time, is probably extremely easy -- and less time consuming in total to set. I disagree the experience will be poor or inconsistent -- those products usually obey a common language of household appliances that works well enough. Physical buttons are usually very responsive and easy to use. There is not much to go wrong.

The ultimate measures here are probably net productivity (time wasted on clock and setting time), cognitive load, and robustness. If you have specific timing requirements maybe the networked clock is for you, but for most people I'd say the good old digital clock is probably still easily best.

Maybe in the future the digital infrastructure will be standardized and simplified enough that more appliances will become connected, but the real value of 'iot-enabling' is still quite low for most objects.

Remember, newer or more flexible isn't necessarily better (even for very complex systems: see UNIX philosophy).

I suspect there's at least some adjustment, which indeed could negate health effects even if they existed (provided the operational adjustment exactly equals their time offset). OTOH If adjustment is incomplete or there's some numerical attachment (universal preference to start at specific hours), then effects could be a possibility.

For personal interactions, you have face-to-face talking and generally (in the context of clustering) access to many people with aligned interests to talk and interact with in person in real time.

Culturally, you have art installations, exhibitions, and all sorts of activities like musical concerts, theater, etc. that aren't exactly reproduced remotely.

In terms of work and hobbies, clusters can give access to tools, workshops, that might be too expensive to have at home. Say using a hackerspace/fab lab vs. having a complete electronics lab in your house or a laser cutter.

All of those should be more readily accessible (in terms of transportation time and cost) if you are well located within a large hub.

Of course, remote does have significant cost benefits in some respects (housing and supplies), and for some professions the benefits of cities aren't strictly necessary, but clustering and the physical world should not be underestimated or thought obsolete.

The first expectation is perhaps that mass unemployment would abide; clearly we don't have mass unemployment in most places (most of the world has a relatively high employment), so the phenomenon must not be occurring or there's the oft-mentioned effect of workers moving to high level tasks.

Since Quantum Supremacy was en vogue, let me coin another term: Robotic Supremacy. That is when a robot (i.e. some kind of automatic machine, not necessarily a humanoid robot) will be able to perform each and every task a human can do more economically. Humans will become unemployable. My hypothesis is that Robotic Supremacy is still far away (maybe a century); and also that this milestone doesn't matter as much as it appears.

The scenario of unemployability is highly unlikely; there's almost always something you can get a human to do that'll pay their food and shelter (we're fantastic machines evolved for billions of years; robots are still far away from domination in many niches). The question is how much this person will earn -- perhaps increasingly less, not more; because it has more competition and the tasks are not as essential as before. So a natural manifestation would be rise in income/wealth inequality. That's precisely what we're seeing. If left unchecked, most places will see a spiral of a tiny elite concentrating all wealth.

First, fantastic that you're doing academic work so early. I believe most students wait far too long to be exposed to this aspect of academia (and their lives), which is far more about asking good questions, achieving deep understanding, and getting good results, than memorizing some procedure that isn't necessarily useful (but happens often in school). Keep creative and keep working on creating a good toolset (find math tools you find interesting/useful and own them!).

Now on to the paper. I'd reiterate the importance of asking good questions almost above results. For example, the result of adversarial sticks and sinks looks good -- but is it asking the right question? If you think realistically, adversarial attacks can occur in a number of ways.

One of them is that a human classifies a dataset one way while a machine another. In this case you would also want the human not to be able to tell you data is weird or there's something funky going on -- that is clearly the case with sticks (and sinks to a lesser extent). A human could be easily trained to spot them, and generally tell something weird is going on.

Another attack scenario is where you can modify some object, like a picture, but have some restriction on how much you can modify it. For example, you can manipulate only some bits of an image, or only perturb a small part of a real-world object that is under classification (say by putting a sticker on a car and fooling a system into thinking it is a dog, or something). If there is no restriction on your perturbation, this problem would be trivial (just replace the data with intended object data). The justification behind sticks and sinks does not look very well fundamented.

So sticks/sinks do not fare too well in either case, despite looking very good in terms of success vs defenses (although there's a chance they could inspire more practical attacks).

The commentary on Haussdorf distance is relevant here, but only on the first case (fooling human judgement), and it is of course an imperfect proxy (the true metric is human perception) -- another hint that fundamentals (and applications) are important to keep in mind.

Overall the paper seems well written and I specially like the numerous illustrations.

Keep the good work and don't forget to always look for the inspiring, beautiful and impactful, and seeking understanding. With a little of this in mind I have no doubt you can achieve very much. Good luck!

Scott's Quantum Supremacy FAQ

Initial ambitions largely set an upper bound on whole project possibilities -- you generally want upper bounds set pretty high. All projects have unknown unknowns that will reveal themselves, and modeling this kind of meta-knowledge is difficult and perhaps not worth the effort.

So to offer a general counterpoint, it may be a valuable skill to listen to constructive advice including ones poking holes in your ideas. The key is to persevere in the face of problems, as long as they're not obviously intransposable (tip: don't go against laws of physics, e.g. thermodynamics or newton's laws). If those hurdles would come up sometime, it might be better to devote more time early on to overcome them.

If you look at it that way, I think the solutions are much more akin to better motivating students, creating better environments (if they don't have them at home).

Motivation is both from showing better job opportunities to be had (higher salaries, employability), the practical value of intellectual and scientific education (to daily life decisions, participating well in democracy -- which cannot work without well educated citizens), and finally an intrinsic value of education -- showing how beautiful intellectual fields are, from a fun approach to mathematics education, to literature, etc.

Environments is making sure the school and classmate interactions are positive, possibly making available study spaces outside regular class time, elective extracurricular activities (programming classes, etc), etc.

Both of those probably go through investing heavily in teacher salaries and adequate teacher education. It's a societal commitment, really.

As long as you clearly define and state things (definitions) and can separate the context-specific [technical] meaning from general usage, I find it's a very good strategy not only for popularization purposes but to inspire and draw some valuable intuition from our daily lives into technical matter.

Naming things in the sciences is an art :)

*: It should be obvious I don't love the usual term 'Kullball-Leibler Divergence' in place of 'Relative Information', although in this case the names are obscure and difficult to pronounce enough to give it an air of rigour and nobility.

https://en.wikipedia.org/wiki/Bird_extinction

The main drivers I think are plain habitat loss, pesticide use (associated decline in insect population), and large scale change in available food and climate.

I think the easiest to address would be pesticide and herbicide use -- widely known to have other negative side effects, such as effect on bees, possible human effects, biodiversity loss, etc. Hopefully it can be replaced by techniques like crop diversification and maybe robotic/biological pest control.

If the room is connected, the is a path from the source A to any chosen point B (by definition). Now take this path and modify it such that it is composed of straight lines, and may touch walls. This new path is a diffuse light path from A to B.

Specular reflection seems much more difficult.

But indeed I'm not an expert by any means on magma properties, perhaps someone else can chime in. If magma is conductive then even less radiation goes through.

Keep in mind at extremely low frequencies the whole Earth is in mid-field or near-field of emissions, so it doesn't behave attenuated like light rays.

To be fair, electromagnetic (i.e. radio) waves do travel at the speed of light through the Earth, though almost none can significantly penetrate the Earth itself, except extremely low frequencies. Electromagnetic interference should be quite easy to measure and look for correlations though; if scientists did their homework (likely), it has been ruled out.