lightsighter

The Trinity test took place in New Mexico, not Nevada.

It feels like there should be some kind of tool analogous to valgrind/ASAN/TSAN/UBSAN for code that is just run on all the images of any kind of paper before it is published to check for these kinds of things. Some of them just seem to be honest mistakes like linking the same image in two different places and this would fix that so nobody needs to bother checking for it in the future. It would also have the byproduct of catching all the malicious image doctoring that some people do which is just horrific for science in general.

We can compare against another mode of human-controlled transportation. There are 1.37 deaths per 100 million passenger-miles driving in the US [1]. In comparison, there are ~0.2 deaths per 10 billion passenger-miles flying. Converting into the same units, there are 137 deaths per 10 billion passenger-miles driving. So you are 685X more likely to die while driving/riding in a car than flying. That's almost three orders of magnitude worse! Humans are pretty terrible drivers in comparison to how good we are at flying.

[1]: https://www.iihs.org/topics/fatality-statistics/detail/state... [2]: https://en.wikipedia.org/wiki/Transportation_safety_in_the_U...

To be fair, it's never been easier to get access to thousands of GPUs in the cloud. It might be expensive, but that is an entirely different kind of barrier. Just a decade ago, it used to be that the only way to get access to thousands of GPUs was to get access to a supercomputer at a national lab. Now anybody with enough money can rent thousands of GPUs (with good interconnects too!) in the cloud. There's certainly a limitation on it from a money perspective, but access to the computational resources themselves is not a problem.

Alright, here's an idea: consciousness is spectrum that occurs when a system develops an automatic self-correcting mechanism for interacting with the external physical universe. In some sense all animals (including humans) wandering on this planet are conscious, because we all learn to build our actions around interactions with the external physical universe, e.g., we learn how to walk/swim/fly under the force of gravity without falling/crashing, we learn that the square peg fits in the square hole and not in the round hole, etc. The feedback from these interactions allows us to automatically adjust our future actions without external help. In this sense we learn what works, aka. what is "true" (at least under the laws of this universe). Some animals happen to have "higher" consciousness in that they interact with the universe in more sophisticated ways, learning "deeper" truths, but all animals possess some degree of consciousness under this definition (my cat is certainly consciousness, she has learned how to manipulate the external world, especially me, perfectly at this point). Consciousness is a matter of degrees, not a binary property that one can satisfy.

This definition also has the nice property of showing why current LLMs don't fit on the spectrum. They don't have any concept of learning what is true and automatically self-correcting. They will happily tell us things that are obviously not true, e.g., the square peg fits in the round hole, and then insist that they are right, when a basic physics experiment will disprove their assertions. Interestingly though, things like linear feedback control systems like we might find in an elevator do possess some degree of consciousness: they interact with the physical world, identify the true position of the elevator, move it where they want, and self-correct when necessary. They might be primitive, but I for one believe that they are certainly "conscious" at some level, and definitely more than LLMs. :)

Almost certainly this definition is incomplete and flawed in many aspects, but I think it's at least self-consistent.

Ironically, in the US, most of the people that want to secede are the ones in states receiving more taxes than they pay.

I'm still using a Pixel 2, now 4.5 years after I got it (lack of security is easy: don't put or do anything I care about on the phone, turn off the power when having sensitive conversations nearby). It can go two days between charging.

The clever thing about interferometers is that they're actually measuring space and time in two different dimensions concurrently and then looking for changes in the interference pattern between light moving along each of the dimensions. Simplistically, imagine a gravitational wave propagating along one dimension of the interferometer (realistically gravitational waves will never be perfectly aligned with any direction of the interferometer). Space will be distorted in that dimension but not the other and we can notice the change in the resulting interference pattern. In practice, gravitational waves will come from all sorts of weird angles, but they will distort each of the two dimensions differently and allow us to figure what direction they were propagating by the interference pattern that's observed.

I never said it stopped all transmission, but it will stop a large fraction of it. We're playing games of probability. Anybody that deals in absolutes is living in a fantasy world.

> What again is the rush to vaccinate the lowest risk age group?

Because they still contribute to chain of transmission to the most vulnerable members of society, likely in a disproportional way since younger members of society tend to socialize more. Fighting a virus is a collective action. In order to stop transmission to the vulnerable members, you need to cut edges along all paths through the graph. Furthermore, additional spread, even among healthy people with no side effects, increases the probability of mutations that lead to more fit variants capable of causing even more sickness and death.

Richard M Stallman: https://en.wikipedia.org/wiki/Richard_Stallman

In my experience, the whole 10 year warranty seems a bit excessive for vehicles that are nigh close to indestructible [1,2,3]. My parents have had multiple Toyotas that made it to 300K-400K miles. My brother's Corolla was rear-ended by a semi doing 30mph and wasn't even close to being totaled. Eight years later and he's still driving it through swamps in south Florida. They are impeccable feats of engineering.

[1] https://www.youtube.com/watch?v=xnWKz7Cthkk [2] https://www.youtube.com/watch?v=xTPnIpjodA8 [3] https://www.youtube.com/watch?v=kFnVZXQD5_k

Please correct me if I'm wrong, but I think all of those system only work inside of a single process. Legion/Regent support distributed multi-node multi-process execution both on supercomputers and in the cloud.

Prior to working on Legion, I worked on a programming language called Sequoia that matches much of what you are describing [1]. In many ways Sequoia was the spiritual ancestor of Legion/Regent.

[1]: http://theory.stanford.edu/~aiken/publications/papers/sc06.p...

The main difference between Legion and TensorFlow is how and when the dataflow graph is constructed. In TensorFlow the graph is constructed lazily (no execution is performed until you've asked for it), it's optimized, and then distributed to processors (GPUs/TPUs) for execution. In Legion, the graph is built, distributed, and executed on the fly. What this means is that Legion can react to things like dynamic control flow (e.g. branches inside of loops) and analyze dependences at runtime to find task parallelism, in a similar way to how your out-of-order CPU extracts instruction level parallelism from a program. Doing things in the TensorFlow model works better when you can see your whole program up front and can "statically" optimize and schedule it because it has lower overheads, but it also has limits to the kinds of programs it can handle; the Legion approach works better when you have dynamic data-dependent behavior in your program and you need to react to it on the fly, but it does have some overhead to the dynamic analysis.

It's worth noting that Regent is the language that implements the Legion programming model. The Legion runtime system is just a C++ library with bindings for C, Fortran, Python, Terra, and Lua. Writing Regent code is much higher productivity than writing to the C++ Legion library directly, but if you want to you can drop down and write your tasks in any of the other languages that Legion supports. You can even mix and match tasks written in different languages.

SF to Amsterdam is 5448 miles, which is 0.29 lightseconds, or 29ms at the speed of light, or about 2 frames of 60fps video, so 5X faster. :)

Except they have no story for building composable libraries in a distributed setting. The story for distributed execution in Julia today is just use MPI, which is a terrible answer. Anyone who has ever use libraries backed by MPI in any language knows that they are inherently not composable. You can't just take an object partitioned across multiple nodes one way by one library and pass it into a second library that expects it to be partitioned a different way. As far as I can tell the Julia language has nothing to say about that, and that makes them a non-starter today for anyone trying to build composable libraries for distributed memory machines.

python: sum([n * n for n in range(1,5,3)])

I had the same thought. The meta-customers here are the insurance companies. If you can prove that the tool improves the performance of doctors so that they get sued less when they use it then the insurance company will give discounts for doctors on their insurance for using it. Presto! We just created a market for doctors to pay for the service.

As someone who has been very laser focused in life I often feel the opposite of you. I finished university, did my PhD, and now have a job doing research in the area of my PhD. I've been working 60-80 hour weeks for the past 10 years. I do some small activities on the side, so life isn't monotonous, but I don't invest in anything nearly as much as I do my work. I often wonder if I'm missing out on life by not committing to more things. So we're on opposite sides of the same coin. The grass isn't necessarily greener...

Actually there is some vision science behind this phenomenon. Your fovea is what you use to focus on things and it's got lots of cone cells that are good at picking up colors and specific details but are not so good at picking up motion. The rest of your eye though has tons of rod cells that effectively see just black and white, but are really good at doing motion detection (mainly to recognize predators and prey in your peripheral vision). So instead of trying to use your fovea to find the plane (which it really sucks at doing), you instead keep your eye moving which exposes much more of the scene to rod cells and let their natural motion detection abilities go to work.

> Smokers and cigarette manufacturers seem to not have some negative third party by your definition

Sure they do. If rates of lung cancer go up because more people smoke, insurance rates for the populace at large will also go up, thereby driving down expendable income for many consumers and depressing economic growth. All of society is the third party when it comes to people making poor decisions about their health.

As an aside, I realize this reasoning can certainly be abused as a justification to make any number of questionable but possibly legitimate behaviors illegal, but it doesn't make it less true in specific cases.

Probably not. The out-of-order execution units in the cores rely heavily on register renaming to remove anti-dependences in the instructions stream and get more instructions executing in parallel. History (so far in most cases) has shown that this is a better approach to getting good performance than having compilers or humans try to statically manage all the physical registers themselves.

It's true that it could happen with any language, but it seems to be especially prevalent and widespread with Fortran. Something about the culture of the language encourages people to dig their heels in and assert that it is good enough.

> And with excellent, state of the art libraries. What’s not to like?

Here's my gripe with Fortran: as a computer scientists that works with combustion chemists, astrophysicists, and nuclear physicists, it's disheartening to see the number of domain specific optimizations that these scientists hand code over and over and over ad nauseam in hundreds of code bases all around the world simply because they think their Fortran compilers are good enough. They don't realize that it would be much more productive if their communities standardized around individual DSL compilers capable of doing higher level transformations specific to their individual domains so they don't have to do complicated transformations by hand (and often mess them up). It's fine if those DSL compilers generate Fortran out the back end so they can interoperate with existing optimized Fortran libraries, but holy hell people, we are living in the 21st century, can we please stop programming in primitive data types (without any unit type checking to boot!) and calling MPI send/recv directly. Scientists have better things to be doing with their time than managing bits directly; raise the level of abstraction by building community specific DSLs, standardize, and let's get on with our lives.

NVLink works fine between GPUs even if the CPUs are x86. Just look at the NVIDIA DGX boxes/stations. https://devblogs.nvidia.com/wp-content/uploads/2017/04/NVLin...

That's all fine and good right up until the next >=10km diameter asteroid has a collision with the Earth, in which case all those humans with cancer will be in the same state as all the humans without cancer. Given that it's been about 65 million years since the last one of these and they tend to happen about every 50 million years or so, you could argue that we are due...

I have a CS PhD from Stanford (graduated 2014) and it is definitely worth it. The thing it actually buys you is freedom to choose your own research, which you can't do with a bachelors or a masters. I now work at an industrial research lab where I get to set my own research agenda within some reasonable bounds of interest of my employer. I pick the projects I work on and am compensated based onto how valuable they are to the company. Importantly failed projects can be just as valuable to the company as successful ones as they can show the product groups what NOT to do. The PhD is effectively a license showing that you've been trained to conduct research. Most companies won't give you the freedom to pick your own research topics without it. For someone that cares a lot about what I'm working on, that freedom is more valuable than money.

All programs are proofs and their results are theorems[1]. If "math is naturally inherent and merely discovered by humans" then so are all programs; programmers are just discoverers and not inventors. Given the proof of Curry-Howard Isomorphism rests on existing maths we can have another discussion about whether mathematical patents should be issued if we don't believe that maths are inherent and are actually invented, but as long as we hold that we can't patent maths then it must also hold that we can't patent programs due to their equivalence. [1]: https://en.wikipedia.org/wiki/Curry%E2%80%93Howard_correspon...