I’ve been using an Oura sleep tracking ring for six months.

In some ways it’s an impressive piece of technology. It’s small enough to not distract me much, and they went overboard in making the user interface simple. Simple, as in there basically aren’t any controls. I just put it on my finger, and occasionally put it on the charger.

Yet it does a poor job of what I expected it to do: track how long I sleep. It occasionally thinks I’m in bed when I’m not wearing it. If I get up to use the bathroom, it’s hard to predict whether it will decide that’s the start or end of my time in bed.

But the Oura reminded me that “8 hours of sleep” isn’t a good description of what I want – that’s just a crude heuristic for “slept well enough that further sleep won’t improve my productivity / health”. The Oura observes other relevant evidence: body temperature, breathing rate, heart rate, and heart rate variability. I.e. things I ignored because they were too hard to evaluate, rather than because I decided they weren’t important.

If I did a strenuous hike yesterday, it will tell me that 7.5 hours of sleep wasn’t enough, whereas if I’d spent yesterday relaxing, it might have told me that 7 hours was plenty, and that I should be ambitious.

It’s somewhat obvious that I need more sleep when a cold raises my body temperature. The Oura convinced me that there’s a much more general pattern of above average body temperature indicating an increased need for sleep.

I’ve tried comparing the Oura’s heart rate variability measurements with those of the emWave2, and I couldn’t see much correlation. I’m inclined to trust the emWave2 more, but I’m not aware of good evidence on the subject.

The Oura also helps track exercise, at least for hiking (it doesn’t seem to do much for weightlifting, but most of my exercise comes from walking/hiking). It reports slightly less calories burned than what I calculate from a cheap Garmin GPS and this calculator. I’m unsure which of those 2 measures is more accurate. If I were only using the GPS to measure calories burned, I’d give up on the GPS, because the Oura doesn’t have problems such as poor reception, or me forgetting to turn it on or off at the start and end of a hike.

It said I slept 3 hours on a red eye flight. My subjective impression was that it was somewhat debatable whether any of that ought to be classified as sleep. But what do I know? I have some evidence that I can sleep without being aware of sleeping (mainly from people reporting that I was snoring, at a time when I thought I was awake and not snoring).

My ring isn’t quite the right size for my ring finger. I ordered it based on prior information about what ring size worked for me, rather than using Oura’s measuring procedure. I’ve ended up wearing on the middle segment of my middle finger instead. That’s works well enough that the difference seems unimportant.

See this comparison with several alternatives for a more detailed analysis.

Mostly, the Oura simply reassured me that I don’t have significant sleep problems, other than the times when it’s obvious that I took too long to fall asleep, or woke up too early. I suspect that the Oura would have been moderately valuable if I had had sleep problems that were hard for me to detect.

Book review: Move Your DNA: Restore Your Health Through Natural Movement, by Katy Bowman.

Move Your DNA does for physical activity what paleo diet advice does for food.

The book is full of suggested movements to practice, making it look somewhat like a yoga book.

Bowman criticizes the common notion of exercise, because it leads to people repeating a tiny set of motions.

Most of us wouldn’t imagine that we had a healthy diet if we ate nothing but carrots, or nothing but liver, even though eating more of those is usually a good idea. Yet plenty of people seem to imagine that they can offset the risks of spending 60 hours per week in chairs by running for a few hours on a carefully maintained surface, repeating a single type of motion with no variation.

A healthier lifestyle would include a wide variety of motion, ideally motivated by the need to accomplish a wide variety of tasks such as carrying wood, digging, pounding nuts, and walking on terrain with lots of little irregularities (she calls this cross-terraining).

How much does it matter? Bowman provides surprising hints, and good theoretical reasons for concern, but leaves me with a good deal of uncertainty about the magnitude of the harm.

Continue Reading

Book review: Human Compatible, by Stuart Russell.

Human Compatible provides an analysis of the long-term risks from artificial intelligence, by someone with a good deal more of the relevant prestige than any prior author on this subject.

What should I make of Russell? I skimmed his best-known book, Artificial Intelligence: A Modern Approach, and got the impression that it taught a bunch of ideas that were popular among academics, but which weren’t the focus of the people who were getting interesting AI results. So I guessed that people would be better off reading Deep Learning by Goodfellow, Bengio, and Courville instead. Human Compatible neither confirms nor dispels the impression that Russell is a bit too academic.

However, I now see that he was one of the pioneers of inverse reinforcement learning, which looks like a fairly significant advance that will likely become important someday (if it hasn’t already). So I’m inclined to treat him as a moderately good authority on AI.

The first half of the book is a somewhat historical view of AI, intended for readers who don’t know much about AI. It’s ok.

Continue Reading

Baze

Convenient, affordable blood tests seem to be important tools for improving health. See my review of The End of Alzheimer’s for hints about why they’re important.

Talking20 and Theranos raised some hopes, then they failed.

Now comes Baze. They shipped me a device that I pressed against my arm. I waited a few minutes, took it off, and shipped it back. I was a bit uncertain about my ability to read the light that changed color to indicate the device had collected enough blood, but I seem to have gotten it right.

Half the time that it took to complete my test involved walking to the nearest FedEx drop box, which is a good deal closer than the nearest LabCorp or Kaiser lab. I had no need to worry about unpredictable delays waiting for a technician to be available to extract my blood.

I got 10 nutrient levels tested for a sale price of about $50. Many of those tests aren’t available on privatemdlabs.com, and the ones that are available are around $50 per nutrient. Life Extension has more of the tests, including Selenium (list price $88, but I normally wait for their spring sale), and Omega-3: $79 for a test that requires me to extract blood from my finger on my own. I had trouble getting enough blood that way, and never got a result, presumably because I didn’t do it well enough. So if I tried to test those 10 nutrients without Baze, I’d have paid maybe $500 and only gotten 7 or 8 results.

Are Baze’s results accurate? I’ve been tested for several of the nutrients previously, and the Baze results for those are similar enough to be reassuring. Their technology seems to have a decent pedigree.

So far, it sounds almost too good to be true. Is there a catch? Maybe. Baze does have a business model that makes me a bit nervous.

Baze is part of Nature’s Way, and tests nutrients in part in order to sell us vitamins in order to correct any deficiencies that it detects.

That does bias Baze away from providing the tests that are most valuable for influencing health-related decisions.

It also biases Baze toward recommending more supplements than is optimal. I don’t see any clear signs that they’re erring in that direction. I also see a distinct shortage of strong arguments in favor of their recommendations.

For vitamin D, they classified my level of 58.3 as excessive, when it’s only about 10% above the level I was aiming for, and there are many people advocating higher levels. That’s a clear sign that they’re not pushing too many vitamins on us.

Yet for choline and omega-3, they classified my blood levels as optimal, yet are still sending me those supplements. There’s at very least something wrong with their explanation here. Yet in both cases, I see some plausible arguments from independent sources that my levels are a bit below ideal, and I had been mildly concerned that I wasn’t consuming enough.

Maybe they’ve got a good explanation hiding somewhere on their blog, but the easy-to-navigate parts of their website are written more for people who want simple and convenient answers from a respected authority. Something feels wrong with their attempt to act like such an authority without providing more evidence of competence than I’ve seen.

They’re also sending me vitamin E and magnesium. I’m pretty confident that I’m consuming a bit more than the RDA for both of those, yet my test results say I’m a bit low in both.

I’m concluding that they have not yet given into the temptation to sell too many vitamins, but they’re putting little effort into reassuring me about this.

Their choice of which tests to do reassured me a bit. They test B12 via methylmalonic acid rather than the less sensitive direct test, and they avoid some nutrients that are more risky to supplement (iron, B6, A, calcium).

Potassium is an important nutrient that many people don’t get enough of. Baze doesn’t do anything with potassium, because potassium supplements are heavily regulated, and because low potassium levels can have causes that ought to be treated by doctors.

Fiber is another important nutrient that many people eat too little of. But that’s rather tricky to evaluate via a blood test – insulin resistance measures say something relevant, but it’s hard to quantify the connection, and doing so might raise novel regulatory issues.

With pretty much all of the nutrients that Baze sells, the evidence for benefits from supplementing is underwhelming, resting mainly on correlations. Where I’ve seen RCTs that test supplementing, only vitamin D seems to show a clear benefit.

I hope Baze focuses more on increasing the variety of biomarkers that it tests for, and less on selling vitamins. I would like to use them for more testing, but not for getting more vitamins. Optimizing our vitamin pill consumption is far from the most valuable goal that this new technology can accomplish.

I suppose the more valuable uses of the technology work best with a fair amount of doctor involvement, and the medical system changes slowly enough that Baze might have needed to introduce the less valuable uses first.

Baze seems good enough now that most people with below-average health (that includes most people over 60) will get a bit of benefit from Baze.

Warning: they report results in different units than I’m used to, so I needed to look up several conversion factors to compare my Baze results to my prior results.

Robin Hanson has been suggesting recently that we’ve been experiencing an AI boom that’s not too different from prior booms.

At the recent Foresight Vision Weekend, he predicted [not exactly – see the comments] a 20% decline in the number of Deepmind employees over the next year (Foresight asked all speakers to make a 1-year prediction).

I want to partly agree and partly disagree.

Continue Reading

Book review: The Paleo Cure, by Chris Kresser.

I wish I had read this when I went paleo 7 years ago. It’s more balanced than the sources I used. Alas, it was published shortly after I finished a big spurt of learning on the subject.

It still has a modest number of ideas that seem new to me, and many ideas that I’d have liked to have known when the book was first published, but which I found through less organized sources.

Continue Reading

Book review: The AI Does Not Hate You: Superintelligence, Rationality and the Race to Save the World, by Tom Chivers.

This book is a sympathetic portrayal of the rationalist movement by a quasi-outsider. It includes a well-organized explanation of why some people expect tha AI will create large risks sometime this century, written in simple language that is suitable for a broad audience.

Caveat: I know many of the people who are described in the book. I’ve had some sort of connection with the rationalist movement since before it became distinct from transhumanism, and I’ve been mostly an insider since 2012. I read this book mainly because I was interested in how the rationalist movement looks to outsiders.

Chivers is a science writer. I normally avoid books by science writers, due to an impression that they mostly focus on telling interesting stories, without developing a deep understanding of the topics they write about.

Chivers’ understanding of the rationalist movement doesn’t quite qualify as deep, but he was surprisingly careful to read a lot about the subject, and to write only things he did understand.

Many times I reacted to something he wrote with “that’s close, but not quite right”. Usually when I reacted that way, Chivers did a good job of describing the the rationalist message in question, and the main problem was either that rationalists haven’t figured out how to explain their ideas in a way that a board audience can understand, or that rationalists are confused. So the complaints I make in the rest of this review are at most weakly directed in Chivers direction.

I saw two areas where Chivers overlooked something important.

Rationality

One involves CFAR.

Chivers wrote seven chapters on biases, and how rationalists view them, ending with “the most important bias”: knowing about biases can make you more biased. (italics his).

I get the impression that Chivers is sweeping this problem under the rug (Do we fight that bias by being aware of it? Didn’t we just read that that doesn’t work?). That is roughly what happened with many people who learned rationalism solely via written descriptions.

Then much later, when describing how he handled his conflicting attitudes toward the risks from AI, he gives a really great description of maybe 3% of what CFAR teaches (internal double crux), much like a blind man giving a really clear description of the upper half of an elephant’s trunk. He prefaces this narrative with the apt warning: “I am aware that this all sounds a bit mystical and self-helpy. It’s not.”

Chivers doesn’t seem to connect this exercise with the goal of overcoming biases. Maybe he was too busy applying the technique on an important problem to notice the connection with his prior discussions of Bayes, biases, and sanity. It would be reasonable for him to argue that CFAR’s ideas have diverged enough to belong in a separate category, but he seems to put them in a different category by accident, without realizing that many of us consider CFAR to be an important continuation of rationalists’ interest in biases.

World conquest

Chivers comes very close to covering all of the layman-accessible claims that Yudkowsky and Bostrom make. My one complaint here is that he only give vague hints about why one bad AI can’t be stopped by other AI’s.

A key claim of many leading rationalists is that AI will have some winner take all dynamics that will lead to one AI having a decisive strategic advantage after it crosses some key threshold, such as human-level intelligence.

This is a controversial position that is somewhat connected to foom (fast takeoff), but which might be correct even without foom.

Utility functions

“If I stop caring about chess, that won’t help me win any chess games, now will it?” – That chapter title provides a good explanation of why a simple AI would continue caring about its most fundamental goals.

Is that also true of an AI with more complex, human-like goals? Chivers is partly successful at explaining how to apply the concept of a utility function to a human-like intelligence. Rationalists (or at least those who actively research AI safety) have a clear meaning here, at least as applied to agents that can be modeled mathematically. But when laymen try to apply that to humans, confusion abounds, due to the ease of conflating subgoals with ultimate goals.

Chivers tries to clarify, using the story of Odysseus and the Sirens, and claims that the Sirens would rewrite Odysseus’ utility function. I’m not sure how we can verify that the Sirens work that way, or whether they would merely persuade Odysseus to make false predictions about his expected utility. Chivers at least states clearly that the Sirens try to prevent Odysseus (by making him run aground) from doing what his pre-Siren utility function advises. Chivers’ point could be a bit clearer if he specified that in his (nonstandard?) version of the story, the Sirens make Odysseus want to run aground.

Philosophy

“Essentially, he [Yudkowsky] (and the Rationalists) are thoroughgoing utilitarians.” – That’s a bit misleading. Leading rationalists are predominantly consequentialists, but mostly avoid committing to a moral system as specific as utilitarianism. Leading rationalists also mostly endorse moral uncertainty. Rationalists mostly endorse utilitarian-style calculation (which entails some of the controversial features of utilitarianism), but are careful to combine that with worry about whether we’re optimizing the quantity that we want to optimize.

I also recommend Utilitarianism and its discontents as an example of one rationalist’s nuanced partial endorsement of utilitarianism.

Political solutions to AI risk?

Chivers describes Holden Karnofsky as wanting “to get governments and tech companies to sign treaties saying they’ll submit any AGI designs to outside scrutiny before switching them on. It wouldn’t be iron-clad, because firms might simply lie”.

Most rationalists seem pessimistic about treaties such as this.

Lying is hardly the only problem. This idea assumes that there will be a tiny number of attempts, each with a very small number of launches that look like the real thing, as happened with the first moon landing and the first atomic bomb. Yet the history of software development suggests it will be something more like hundreds of attempts that look like they might succeed. I wouldn’t be surprised if there are millions of times when an AI is turned on, and the developer has some hope that this time it will grow into a human-level AGI. There’s no way that a large number of designs will get sufficient outside scrutiny to be of much use.

And if a developer is trying new versions of their system once a day (e.g. making small changes to a number that controls, say, openness to new experience), any requirement to submit all new versions for outside scrutiny would cause large delays, creating large incentives to subvert the requirement.

So any realistic treaty would need provisions that identify a relatively small set of design choices that need to be scrutinized.

I see few signs that any experts are close to developing a consensus about what criteria would be appropriate here, and I expect that doing so would require a significant fraction of the total wisdom needed for AI safety. I discussed my hope for one such criterion in my review of Drexler’s Reframing Superintelligence paper.

Rationalist personalities

Chivers mentions several plausible explanations for what he labels the “semi-death of LessWrong”, the most obvious being that Eliezer Yudkowsky finished most of the blogging that he had wanted to do there. But I’m puzzled by one explanation that Chivers reports: “the attitude … of thinking they can rebuild everything”. Quoting Robin Hanson:

At Xanadu they had to do everything different: they had to organize their meetings differently and orient their screens differently and hire a different kind of manager, everything had to be different because they were creative types and full of themselves. And that’s the kind of people who started the Rationalists.

That seems like a partly apt explanation for the demise of the rationalist startups MetaMed and Arbital. But LessWrong mostly copied existing sites, such as Reddit, and was only ambitious in the sense that Eliezer was ambitious about what ideas to communicate.

Culture

I guess a book about rationalists can’t resist mentioning polyamory. “For instance, for a lot of people it would be difficult not to be jealous.” Yes, when I lived in a mostly monogamous culture, jealousy seemed pretty standard. That attititude melted away when the bay area cultures that I associated with started adopting polyamory or something similar (shortly before the rationalists became a culture). Jealousy has much more purpose if my partner is flirting with monogamous people than if he’s flirting with polyamorists.

Less dramatically, We all know people who are afraid of visiting their city centres because of terrorist attacks, but don’t think twice about driving to work.

This suggests some weird filter bubbles somewhere. I thought that fear of cities got forgotten within a month or so after 9/11. Is this a difference between London and the US? Am I out of touch with popular concerns? Does Chivers associate more with paranoid people than I do? I don’t see any obvious answer.

Conclusion

It would be really nice if Chivers and Yudkowsky could team up to write a book, but this book is a close substitute for such a collaboration.

See also Scott Aaronson’s review.

[I have medium confidence in the broad picture, and somewhat lower confidence in the specific pieces of evidence. I’m likely biased by my commitment to an ETG strategy.]

Earning to Give (ETG) should be the default strategy for most Effective Altruists (EAs).

Five years ago, EA goals were pretty clearly constrained a good deal by funding. Today, there’s almost enough money going into far future causes, so that vetting and talent constraints have become at least as important as funding. That led to a multi-year trend of increasingly downplaying ETG that was initially appropriate, but which has gone too far.

Continue Reading

There are a number of investment ideas that pop up about once per generation, work well for years, and then investors get reminded of why they’re not so good, and they get ignored for long enough that the average investor doesn’t remember that the idea has been tried.

The idea I’m remembering this month is known by the phrase Nifty Fifty, meaning that there were about 50 stocks that were considered safe investments, whose reliable growth enabled investors to ignore standard valuation measures such as price/earnings ratios, dividend yields, and price to book value.

The spirit behind the Nifty Fifty was characterized by this line from a cryonaut in Woody Allen’s Sleeper (1973): “I bought Polaroid at seven, it’s probably up millions by now!”.

There was nothing particularly wrong with the belief that those were good companies. The main mistakes were to believe that their earnings would grow forever, and/or that growing earnings would imply growing stock prices, no matter how high the current stock price is.

I’ve seen a number of stocks recently that seem to fit this pattern, with Amazon and Salesforce mostly clearly fitting the stereotype. I also ran into one person a few months ago who believed that Amazon was a good investment because it’s a reliable source of 15+% growth. I also visited Salesforce Park last month, and the wealth that it radiated weakly suggests the kind of overconfidence that’s associated with an overpriced stock market.

I took a stab at quantifying my intuitions, and came up with a list of 50 companies (shown below) based on data from SI Pro as of 2019-09-20, filtered by these criteria:

  • pe_ey1 > 30 (price more that 30 times next year’s forecast earnings)
  • mktcap > 5000 (market capitalization more than $5 billion)
  • prp_2yh > 75 (price more than 75% of its 2 year high)
  • rsales_g5f > 50 (5 year sales growth above the median stock in the database)
  • sales_y1 < 0.33333*mktcap (market capitalization more than 3 times last year’s sales)
  • yield < 3 (dividend yield less than 3%)
  • pbvps > 5 (price more than 5 times book value)
  • epsdc_y2 > 0 (it earned money the year before last)

I did a half-assed search over the past 20 years, and it looks like there were more companies meeting these criteria in the dot com bubble (my data for that period isn’t fully comparable), but during 2005-2015 there were generally less than a dozen companies meeting these criteria.

The companies on this list aren’t as widely known as I’d expected, which weakens the stereotype a bit, but otherwise they fit the Nifty Fifty pattern of the market seeming confident that their earnings will grow something like 20% per year for the next decade.

There were some other companies that arguably belonged on the list, but which the filter excluded mainly due to their forward price/earnings ratio being less than 30: BABA (Alibaba Group Holding Ltd), FB (Facebook), and GOOGL (Alphabet). Maybe I should have used a threshold less than 30, or maybe I should take their price/earnings ratio as evidence that the market is evaluating them sensibly.

This looks like a stock market bubble, but a significantly less dramatic one than the dot com bubble. The market is doing a decent job of distinguishing good companies from bad ones (much more so than in the dot com era), and is merely getting a bit overconfident about how long the good ones will be able to maintain their relative quality.

How much longer will these stocks rise? I’m guessing until the next major bear market. No, I’m sorry, I don’t have any prediction for when that bear market will occur or what will trigger it. It will likely be triggered by something that’s not specific to the new nifty fifty.

I’m currently short EQIX. I expect to short more of these stocks someday, but probably not this year.

ticker company pe_ey1 mktcap sales_y1 yield pbvps
AMT American Tower Corp 52 100520 7440.1 1.7 18.21
AMZN Amazon.com, Inc. 54 901016 232887 0 16.67
ANSS ANSYS, Inc. 31.8 18422.3 1293.6 0 6.46
AZPN Aspen Technology, Inc. 30.5 8820.8 598.3 0 22.2
BFAM Bright Horizons Family Solutio 37.2 9181.8 1903.2 0 10.21
CMG Chipotle Mexican Grill, Inc. 48 23067.9 4865 0 15.04
CRM salesforce.com, inc. 50.1 134707 13282 0 7.02
CSGP CoStar Group Inc 49.3 21878.4 1191.8 0 6.74
DASTY Dassault Systemes SE (ADR) 32.9 37683.3 3839 1 7.05
DXCM DexCom, Inc. 110.3 14132.9 1031.6 0 20.44
EQIX Equinix Inc 70.2 48264.7 5071.7 1.7 5.46
ETSY Etsy Inc 61.7 7115.6 603.7 0 16.42
EW Edwards Lifesciences Corp 36.7 44736.3 3722.8 0 13.06
FICO Fair Isaac Corporation 36.5 9275.5 1032.5 0 33.71
FIVE Five Below Inc 33.6 7152.1 1559.6 0 10.86
FTNT Fortinet Inc 31.6 13409.2 1801.2 0 11.93
GDDY Godaddy Inc 67.2 11976.7 2660.1 0 12.41
GWRE Guidewire Software Inc 70.3 8835.9 652.8 0 5.84
HEI Heico Corp 49.3 15277.3 1777.7 0.1 10.58
HUBS HubSpot Inc 94.6 6877.4 513 0 10.93
IAC IAC/InterActiveCorp 38.3 19642.5 4262.9 0 6.51
IDXX IDEXX Laboratories, Inc. 49.3 23570.6 2213.2 0 138.03
ILMN Illumina, Inc. 44.3 44864.4 3333 0 10.5
INTU Intuit Inc. 31.6 70225.1 6784 0.8 18.67
INXN InterXion Holding NV 106.8 5995.1 620.2 0 7.95
ISRG Intuitive Surgical, Inc. 38.8 61020.9 3724.2 0 8.44
LULU Lululemon Athletica inc. 33.7 25222.7 3288.3 0 16.36
MA Mastercard Inc 30.1 276768 14950 0.5 55.23
MASI Masimo Corporation 41.8 8078.9 858.3 0 7.8
MDSO Medidata Solutions Inc 43.4 5734.1 635.7 0 8.35
MELI Mercadolibre Inc 285.4 27306.2 1439.7 0 12.59
MKTX MarketAxess Holdings Inc. 56.7 12941.1 435.6 0.6 18.93
MPWR Monolithic Power Systems, Inc. 31.5 6761.2 582.4 1 9.44
MTCH Match Group Inc 38.4 22264.7 1729.9 0 105.51
OLED Universal Display Corporation 45.5 8622.8 247.4 0.2 11.36
PAYC Paycom Software Inc 51.3 12812.8 566.3 0 28.6
PCTY Paylocity Holding Corp 47.6 5150.5 467.6 0 17.01
PEGA Pegasystems Inc. 167.2 5686.4 891.6 0.2 10.14
PEN Penumbra Inc 134 5142.8 444.9 0 11.3
RMD ResMed Inc. 30.6 19181.5 2606.6 1.2 9.33
RNG RingCentral Inc 139.4 11062.4 673.6 0 30.93
ROL Rollins, Inc. 43.6 11383.4 1821.6 1.2 15.04
RP RealPage Inc 31.3 6084.5 869.5 0 5.3
TAL TAL Education Group (ADR) 39.4 21326.2 2563 0 8.45
TECH BIO-TECHNE Corp 34.6 7483.8 714 0.6 6.52
TREX Trex Company Inc 30.2 5074.2 684.3 0 13.05
TYL Tyler Technologies, Inc. 43.5 10004.5 935.3 0 6.98
VEEV Veeva Systems Inc 62 21366.2 862.2 0 15.27
VRSK Verisk Analytics, Inc. 32.3 25948.4 2395.1 0.6 11.73
ZAYO Zayo Group Holdings Inc 42.5 7997.8 2578 0 5.95