tvural

Learning new languages in-depth is an incredibly taxing thing, because there are whole jigsaw pieces of syntax you have to learn at once. But worth doing the right way if you have time.

It's worth reading the Ray Dalio debt crisis book to see how this might end - the most likely scenario is the US, Japan, and EU all encountering cash flow problems in the next recession that they resolve by devaluing their currency. In that sense people have less money today than they believe. If all the major currencies and stock markets crash at the same time, there's no safe place for everyone to keep their money.

I think the decline of Silicon Valley will have much more to do with the end of Moore's Law than any kind of social or political trend. The name is very predictive - if there will be nothing left to do with Silicon, maybe there will be nothing left to do in Silicon Valley, and the computer industry generally.

Dropbox Paper has worked quite well for me.

Financial projections will undervalue Tesla because they are very different from other car companies - they are a monopoly with gigantic barriers to entry on many fronts. Other carmakers could copy some features of Tesla's cars, but they can't easily copy the multi-billion-dollar factories that make other features possible. They have better engineers than other car companies and a culture that works them harder. They avoided the dealership system and built their own stores. Elon Musk's PR stunts and Twitter feed are a much more effective method of advertising than the TV and Internet ads other car companies use.

True, but it can be a significant problem for legitimate companies. After going public Tesla had the highest short interest of any major stock at 72%. The other companies that were in the top 5 at the time are all bankrupt or close to it now. Short sellers perceived it as an unrealistic joke, and tried to convince everyone else of that, which could easily have become self-fulfilling.

I would also point out that short sellers and "activist investors" are a significant downside to being public. Essentially you give investors focused on these strategies a financial incentive to destroy the company's long-term value. Activism has been a big problem for drug companies where investors will try to fire all the scientists so that profits will go up for a few years.

Healthcare is a much bigger problem for the government than student loans. The federal government spends about 1 trillion per year on healthcare, while the total of all student debt in the U.S. is around 1 trillion. This is true on a societal level too. U.S. healthcare spending is 17% of GDP, up from 5% in the 1960s. The problem with student loans is that the burden falls mostly on the least financially secure people. This is as opposed to Medicare/Medicaid where the government tries to foot the bill rather than allow people to get in massive debt.

I predict that BMIs are going to suffer from the same problem as AI, where the applications that are working in the short-term get very overestimated because they are confused with the long-term where you create a singularity. If you had a BMI that could read/write the entire brain on neuron-level resolution, you could create computer back-ups of people, and if hardware were fast enough you could create superhuman intelligence. If you just have cochlear implants and prosthetics, the best case is a world where nobody is impaired, which is good, but still very far from a singularity. The Neuralink version is that if you can do telepathy, that might be valuable in some situations, but it will probably just be like faster email until the computers become smarter than us.

I think it's unclear how much progress has been made on the superhuman AI problem. We haven't pinned down a good definition of intelligence, or figured out what it is that makes monkeys smarter than mice and us smarter than monkeys. We do have a lot of progress in specific domains, like image/speech recognition, but it's hard to tell whether they're on the critical path to superhuman AI because we don't know what the critical path is yet. That makes the timeline unpredictable, but biased a priori towards "far away". It's possible that better hardware will accelerate progress, but with CPU clock speeds flattening out, significantly better hardware is not guaranteed in the future.

$200M usually isn't enough for even one drug - $500M is on the low end, and not all drug development works out. So it's plausible that the money is better allocated here than in treatments.

But also, from the article: "Based on some of 23andMe’s newer hires, we wouldn’t be shocked to see the company dive into drug development, either."

There are likely some ulterior motives behind the way cryptocurrency is currently being covered. If you have sway in a news organization, it would be quite lucrative to invest in some crypto-related companies, and then encourage all stories written about cryptocurrency to take a positive view. The persuasion works best when it's hidden, like a story about a teenager becoming rich by buying into bitcoin early. This is fine, but only if all such stories come with a disclosure. Otherwise it's an ethically challenging position to be in, like the rating agencies who kept quiet to avoid losing their clients during 2008.

This link isn't behind a paywall: https://techcrunch.com/2017/08/22/this-23-year-old-just-clos...

Their economy is based on copying things that worked in the U.S. In the 1980s people were worried based on Japan's rate of growth that it was going to overtake the U.S. Instead they just ran out of things to copy, and the growth completely flattened. At some point you can only grow by doing new things, and for whatever reason Japan has historically been very bad at that.

This reminds me of Robin Hanson's criticism of medical spending - exercise and diet are much better predictors of health than money spent on healthcare, meaning that there's a large amount of waste in the system. I suspect something similar here, the amount of money people spend on education is uncorrelated from what they get out of it. The other factors, like whether you went to a top school and what you did in college dominate the ROI. There is a tendency to treat education as an insurance policy, where all you have to do is buy in, and you are somehow protected from falling through the cracks in society. It seems that college as an insurance policy has stopped working if it ever did work, and it exposes a fact that's somewhat uncomfortable, you can't just pay these colleges money to have your career set, you have to figure out what to do yourself.

There's definitely more that can be done behaviorally - the solution to obesity today probably looks more like eating better than taking diet pills. But this is the anti-technological solution that can only be pushed so far, like trying to solve the energy crisis by telling everyone to drive smart cars.

Also, there's often nothing you can do behaviorally. Even people with near perfect behavior will eventually get Cancer or Alzheimer's or Heart Disease. Most of the drugs approved to treat these diseases are something like band-aids that decrease your risk 10% or allow you to live a couple months longer. Maybe too many people are making drugs that work barely well enough for the side effects to be worth it. But almost anything that can get past the FDA will work economically, since there are usually very few alternatives from the buyer's perspective.

Negative press like this probably contributes to the FDA being so conservative. They get none of the benefits when a great new drug is approved, but get scapegoated when they approve something dangerous even to small numbers of people. So they just approve as few drugs and medical devices as they can get away with - it's a massive misalignment of incentives.

One might think drugs would be more dangerous if the FDA were less conservative, but I suspect the opposite. If it takes you ten years and 2 billion dollars to release a new drug, and then you find out it has some terrible side effect in 0.1% of the population, you don't get to fix the problem. You might not withdraw it either, since the drug would still be a net good for society. And if one drug is approved instead of ten, patients are forced to stick to the flawed solution.

The far bigger problem than drugs with side effects is a lack of drugs. Around 1/3 as many drugs get approved today relative to 1970, and the costs of a drug approval have risen 5 times faster than inflation since then. Medicine has been an anti-technological field for the last few decades - it's doing less with more. The most valuable biomedical companies are the ones that have been around since the 1800s - one could imagine what computers would look like today if IBM were still the largest IT company.

> "AI's don't need to learn to code any more than they need to learn to use photoshop. They need to learn to provide functionality (or in this case manipulate image data)."

This is interesting. My counterpoint would be that if you rely on AI over programs you lose human-editability and determinism. So fixing a bug or adding a new feature might mean diving into some opaque model rather than adding a few lines of code. You couldn't do anything where consistency is important, like security, manipulating a database with important information, or GUI design. I think that at least protects large swaths of software development.

Even this example seems less like a replacement for Photoshop and more like a cool new feature Photoshop could add

"Fix those two problems, and a new search engine could be better than Google. Whether anyone would notice is questionable."

Yeah, I don't think many people will care about the difference between good and perfect. You might be able to find a niche in search that Google is ignoring, but you would have a hard time expanding from there into general search.

In that sense Google is a bet against technology - you would invest in Google if you believe the Web's going to stay the same for a long time and nothing will replace it.

One possibility is that the resolution of VR goggles becomes so good that you can simulate looking at a screen. Then you might have something as easy to carry as a smartphone but with unbounded screen size.

Care to elaborate?

If you're worried AI is having no impact at all, some of the places I can think of where it's already had an impact are Google Search, Palantir, and driver assistance (as in Tesla's Autopilot).

So the tools we have now aren't total frauds, and there's probably more to be discovered.

What all three of those examples have in common is that they involve an AI assisting human decision-making. I think that this will be the most lucrative area in the future as well. If you have a set of data a human can't even look at, like a million web pages, the AI doesn't have to be that good at processing it to be useful, since it has no competition from humans. On the other hand, it's rare that you ever need to come up with a million new concepts - you probably just need one. So humans easily outcompete AI at coming up with new ideas. Maybe it's a bit vague what "new ideas" means, but certainly if you're capable of generating new ideas, then you're at least generating turing-complete outputs. Nobody is even trying to do that right now. In the case of image recognition, there are some complex ideas needed to make it work, but at the end you're just outputting a label, getting your model to use the most basic DSL possible.

First, here is a link that is not hidden behind a paywall: https://arxiv.org/pdf/1607.04878v2.pdf

> there is substantial evidence supporting a holographic explanation of the universe—in fact, as much as there is for the traditional explanation of these irregularities using the theory of cosmic inflation.

This is a bit misleading, especially the phrase "substantial evidence". I bet that the authors of the paper would not have used this phrasing. From the paper:

> We emphasise that the application of holography to cosmology is conjectural, the theoretical validity of such dualities is still open and different authors approach the topic in different ways.

Essentially, their paper shows that a holographic model cannot be ruled out simply by comparing the predictions it makes for the CMB to observation. It also gives some intuition for why a holographic model might make sense - at sufficiently early times in the Universe quantum and gravitational effects begin to coincide, and in other contexts people have modeled quantum gravity using "a quantum field theory with no gravity in one dimension less". The paper finds, however, that there is no empirical case to be made for discarding the standard model of inflation:

> We see that the difference between evidence for [the standard model] and HC predictions is insignifcant, with marginal preference for HC, depending on the choice of priors.

In case anyone else didn't know, this is what the title references:

https://en.m.wikipedia.org/wiki/First_they_came_...

People are worried about AI risk because ensuring that the strong AI you build to do X will do X without doing something catastrophic to humanity instead is a very hard problem, and people who have not thought much about this problem tend to vastly underestimate how hard it is.

Whatever goals the AI has, it will certainly be better at achieving them if it can stay alive. And it will be more likely to stay alive if there are no humans around to interfere. Now you might say, why don't we just hardcode in a goal to the AI like "solve aging, and also don't hurt anyone"? And ensure that the AI's method of achieving its goals won't have terrible unintended consequences? Oh, and the AI's goals can't change? This is called the AI control problem, and nobody's been able to solve it yet. It's hard to come up with good goals for the AI. It's hard to translate those goals into math. It's hard to prevent the AI from misinterpreting or modifying its own goals. It's hard to work on AI safety when you don't know what the first strong AI will look like. It's hard to prove with 99.999% certainty that your safety measures will work when you can't test them.

Things will not turn out okay if the first organization to develop strong AI is not extremely concerned about AI risk, because the default is to get AI control wrong, the same way the default is for planets to not support life.

My counterpoint to the risks of more limited AI is that limited AI doesn't sound as scary when you rename it statistical software, and probably won't have effects much larger in magnitude than the effects of all other kinds of technology combined. Limited AI already does make militaries more effective, but most of the problem comes from the fact that these militaries exist, not from the AI. It's hard for me to imagine an AI carrying out a military operation without much human intervention that wouldn't pose a control problem.

--------- Edited in response to comment--------

Matrices are linear maps, and matrix multiplication composes the linear maps. Now statements like the determinant of the product is the product of the determinants, the trace is preserved under similarity transforms (since similarity transforms just rewrite the linear map in a different basis), etc. are intuitive.

I've always felt that these explicit calculations don't really get to the point. You can memorize them and still not really understand what's going on.

The best explanation is probably that squared error gives you the best fit when you assume your errors should normally distributed.

Things like the fact that squared error is differentiable are actually irrelevant - if the best model is not differentiable, you should still use it.

The difference between this and the Iphone 1 is that the technology is not there yet for fully autonomous cars, and it's unclear you can get there just by trying really hard and spending lots of money. Fully self-driving cars wouldn't just have to stay in their lane on the freeway, they would have to make the decisions during edge-case driving situations that a human would. That means knowing to slow down when a ball rolls across the road, reading people's hand signals, responding correctly when a police car tries to stop traffic on the freeway, etc. I'm not sure how you even begin to handle cases like that. Our driving situation was designed for humans, so you would have to make the algorithm "think like a human" and understand all our wierd idiosyncrasies. So far, machine learning has had very little luck with trying to similate human decision-making.

What Tesla has done so far with "autopilot" is a big deal, but releasing a "video of a car driving itself" is the classic marketing trick with machine learning, where you cherry pick aituations in which your system does well, and most people don't second guess it.

For coding specifically, I find that I procrastinate much less when I'm debugging. Having the compiler give me bite-sized, well-defined tasks to complete can be very good for morale. So I try to avoid writing code for more than an hour or so without testing some part of it.

Thiel might have ulterior motives, but his support of Trump is not too surprising if you've listened to his other ideas. He feels that the political agenda has been overrun by distractions, like "who can use which bathroom", at the expense of real problems, like the stagnation of median wages over the last forty years. He prefers the days when the national conversation was about how to beat the Soviets, and sees the lack of substance in our political agenda today as tied to a slowdown in progress. So he probably prefers Trump's speeches about China and decaying inner cities to Clinton's about fairness and diversity.

He believes that most of Silicon Valley is naive politically, and that the popularity of social liberalism there is just a moral fashion. He is a liberarian, and believes that the Valley's instincts are libertarian, not liberal. He has a slightly pessimistic outlook on the future, and believes that America has been falling behind since 1969, "when Woodstock started... and the hippies took over the country". That aligns well with the central point of Trump's campaign - America has started losing and we need to "make it great again".

He's often said that one of his favorite interview questions is, "tell me something you believe to be true but which nobody agrees with you on". His support of Trump falls into that category. 40% of the population agrees with him, but the people closest to him see his opinion as unthinkable. He seems to take pleasure in having opinions like that. He has said that he believes climate change is "more pseudoscience than science", roughly agreeing with Trump. Thiel backed up his position by saying "whenever you can't have a debate, I often think that's evidence that there's a problem".

He likely sees the stories that have come out against Trump recently as worth ignoring when the future of the country is at stake, and Trump is the only candidate who can focus the national agenda on the right issues.