beisner

authoritarian is probably the right word, rather than left/right.

But we are all computers made of DNA tho

There are quantum-resistant encryption schemes.

See: https://en.wikipedia.org/wiki/Post-quantum_cryptography

parent would probably say, "grid-scale battery storage". or natural gas.

No, they own both an asset and a liability, secured by that asset.

As I understand it, I think that any randomly generated SAT problem will not be able to be solved quickly; however, SAT problems in the real world, even when they're large, tend to be quick for SAT solvers (most people solve the "Wedding Seating" problem for their own weddings, after all!), Why ordinary SAT problems tend to be easy to solve, last I checked was an open problem.

Gotcha, understood. Mirroring @Denzel's edit, I think that rationale makes sense for configuring some of the intermediate glue which is project-specific. For instance, running test suites, orchestrating deployment, and other more custom actions which aren't terribly reusable across projects (think Bazel's 'genrule'). Caching test results in a distributed way is a great use case, and environment isolation (i.e. without forcing the user to define a Dockerfile or similar, and leaving environment configuration to the build system) seem like great use cases.

However, I think for describing actual build toolchains this way of implementing things might end up being significantly more complicated (and require even more arcane end-user cognitive load) than the Starlark/Python-based build rule / toolchain / constraint approach for actually assembling libraries, binaries, and other compiled artifacts. There are a combinatorial number of backends, conditional options settings, etc. which will be hard to capture with a purely declarative system. For instance, a C++ binary might needs platform-specific compiler flags, or some #ifdef nonsense. YAML doesn't have a clean way of implementing conditionals based on some constraint. So for heterogenous ecosystems (C/C++, Python, container assembly, GPU development, microcontrollers, etc.) this pushes a ton of complexity into the build rules themselves which may be opaque to end users (and thus introduce a ton of cognitive load, steep learning curves, hard-to-debug errors, etc.).

What's the motivation for using YAML instead of Starlark (Bazel, Buck) or something closer to Python (Pants, please.build)? Seems as though much of the other monorepo tools have (kinda) standardized on this.

As I understand it, the primary reason these build systems leverage these Python-variants is so that the build rules, toolchains, constraints, and build definitions can all be written in the same language (since build rules often require some programmatic behavior). Perhaps with a future vision of them being totally interoperable across build systems.

Real estate is priced at the margins, because it's rather illiquid. Here's an estimate, there are about 10M people living in Portugal, maybe 5M living in areas (cities, coasts) where wealthy foreigners might want to settle. Average 3 people per household, that's 1.6M homes. Housing vacancy rates in desirable areas are usually 1-3%, so maybe 15-45k available homes at any given time in desirable areas. An influx of a few thousand extra wealthy people (say 3-5k, not just Americans) in a short period of time could certainly have an impact on current vacant housing prices, which might last a good number of years (prices tend to go up quickly, and down slowly).

In the US, many people drive 5+hr long-distance trips a few times a year at least (visit relatives, vacation, etc). So the idea of either getting fleeced or stranded a few times a year is unappealing.

that's profit after employee compensation costs.

I think it's more of a sarcastic / cynical take - kind of a "you reap what you sow" sentiment. This IS what capitalism is, with a weak social safety net and without controls on the accumulation of stupid levels of capital. If you don't believe that the people who invested in Coinbase should have had the ability to fling around billions of dollars on a whim, then it's difficult to feel bad when they spend it poorly.

BNPL usually breaks purchases up into discrete monthly payments up-front at checkout. Whereas your credit card lumps it together into one balance, which you can pay over time.

It's probably reasonably straightforward for well-connected VC people to raise $1bn from philanthropic rich people, which might have seemed like enough money at the time to make progress on the problem of AGI. But I suspect they've realized that the actual amount of capital required to cover headcount (some of the most expensive employees in the world) and compute (some of the largest specific+expensive compute requirements out there) that would be necessary to advance the state of the art far enough to create a real product that might be cash-flow-neutral is probably on the order of $10-30bn. And that's a tremendous amount of money for a philanthropist or set of philanthropists to sink into, far more than even the most well-connected people can marshal in donations from American billionaires. And the government likely wouldn't funnel that kind of money into an organization that has only a few hundred people. So the only remaining route (or at least the easiest for a bunch of established capitalists) is to raise money as a for-profit venture.

People who travel a lot for work are often extremely attuned to the credit card rewards game, so getting only 1% back is unrealistic. You can eek out at least 5% back (and often much higher, depending on redemptions) on all travel expenses. So in this(extremely inexpensive, bare-bones) travel scenario it’s more like $65. With international or frequent travel, it’s not hard to arrive at a couple extra thousand dollars a year.

Very curious to see what this means for ROS.

While true, there’s a relatively small upper bound on how many bits of information are in this pre-training. Specifically, in the form of how much information is contained in DNA, which is only a couple gigabytes.

Alongside simple banking, this is a service that should be offered, for free, by the US Post Office.

Sure; my point is that both perspectives are equally arbitrary. Neither drive me nuts.

On the flip side, the child didn't consent to being born, and indirectly any suffering a child endures can be traced back to a parent's decision to bring them into the world.

The idea of having a public record that attests to when an event happened is interesting, although not sure it has to be blockchain for it to be useful.

Each sensor could have a unique cert

Sure, there are of course runtime checks you can do (and I now do after having been bitten by this before), but that’s not necessarily better than having a static analyzer which can guarantee correctness without executing a single line of code.

So there are information theoretic ways to certify that media is genuine, if you assume trust at least somewhere in the process. Basically just cryptographic signing.

For instance, a camera sensor could be designed such that every image that is captured on the sensor gets signed by the sensor at the hardware level, with a certificate that is embedded by the manufacturer. Then any video released could be verified against a certificate provided by the manufacturer. Of course, you have to trust the manufacturer, but that’s an easier pill to swallow (and better supported by our legal framework) than having to try and authenticate each video you watch independently.

There are issues that can arise (what if I put a screen in front of a real camera??, what if the CIA compromises the supply chain???), but at the end of the day it makes attacks much more challenging than just running some deepfake software. So there are things that can be done, we’re not destined for a post truth world where we can’t trust any media we see.

It's an optional typechecking tool. Speaking as an ML researcher who has authored my fair share of linear algebra bugs due to broadcasting, being able to optionally typecheck specific critical regions of the codebase would be a killer feature.

Variadic Generics has the potential to enable all the scientific computing libraries to provide really nice type hints. One step closer to eliminating insidious broadcasting bugs… or running training for 10 hrs only to have a validation loop fail because of a shape mismatch.

In the US, a postdoc at a top CS school makes like $60-80k and works 60-80hrs a week, whereas a new grad of the same caliber with just a bachelors (think top 10% of graduating class) can reasonably expect to make >$200k working 40 hours a week at a FAANG. Meanwhile, Research Scientist roles (the common alternative to doing a postdoc in CS) recently cracked $400k starting. These are numbers I have personally experienced. (Verifiable on aipaygrad.es and levels.fyi).

CS is certainly an outlier in terms of salaries, but ANY person capable of landing a postdoc in a STEM field at a competitive school in the US is capable of figuring out a way to get a job as a software engineer in Big Tech (the grind of leet code is nothing compared to the grind of doing a PhD, full stop). This is why >60% of PhD grads in CS across top schools are ending up in industry, and I knew a TON of PhDs from physics, math, chemistry, etc. who left research immediately after graduating and are now SWE.

That’s how Bazel Python works

Those things aren’t necessary when good-faith communication can be assumed. This is an Internet forum, so good-faith communication cannot be assumed.

There are extreme barriers to entry for airplane manufacturing. You can’t just get a couple million dollars from some VC and have an MVP 747 in 6 months. There’s a reason that only 2.5 (.5 being COMAC) companies in the world that produce commercial airliners. A new player has to have tens of billions of dollars in investment and also be willing to wait 10-15 years before a single plane is approved by the FAA and other regulatory bodies, and unless you have a radically different airframe (Boom, etc) your company will directly compete on price with Boeing and Airbus, who have a 100 year head start. There’s also a strong incentive to not let Boeing go under, because 1.5 airline companies is dramatically worse for [safety, cost, innovation] than having 2.5.

So not only is it NOT a free market (MASSIVE long-term investment plus enormous warranted regulatory burden), but we don’t want it to be one! Because we don’t ever ever want Boeing or Airbus to fail because that would be devastating to national security and safety etc.

All this is to say that “a race to the bottom” in this context is maybe more harmful than in other markets, because there really isn’t much backpressure (outside regulatory oversight) to keep standards high.