I have type 1 diabetes, an autoimmune condition which requires I constantly measure my blood sugar and use insulin to control it. The simplified model is: blood sugar goes up as a result of food and as a result of time, down as a result of insulin, and the goal is to keep it inside a narrow range. I have a Dexcom G4 continuous glucose sensor, which measures blood sugar every five minutes, and a a Medtronic insulin pump, which gives insulin on command; but until recently, there was no way to make these talk to each other, so I had to make all insulin-delivery decisions by hand.

OpenAPS is an open source project which takes continuous glucose monitor data, and uses an insecure 900MHz radio protocol to command the pump. It works, and it works very well! One thing I found left a lot to be desired, though, was the case hardware–none of the options people had designed were both sufficiently protective, and small enough to fit in my pocket. (While I was waiting for the case to be printed and to arrive, I broke the USB OTG connector off my Explorer Block, by keeping it in my pocket with only an antistatic bag for protection, thus demonstrating that a good protective case is indeed necessary). So I downloaded the trial version of Fusion 360, measured all the parts carefully, and designed one. This post documents the case, what parts it works with and how to assemble an OpenAPS rig using it if you are using the right components (in particular, an Intel Edison and a Dexcom G4).

The rest of this post is aimed at people who use OpenAPS or are looking to get started with OpenAPS, who arrived here via a link from the “Get your rig hardware” section of the OpenAPS documentation. It is probably not as interesting to others.

Parts:

• The 3D printed plastic case
• A Dexcom G4 receiver
• Intel Edison
• 900MHz Explorer Block
• 10cm right-angle USB OTG cable
• Two USB battery packs with builtin AC
• Retractable USB cable
• Black electrical tape

The end result is pocket sized (albeit just barely), and fairly close to the theoretical limit of how small it can be without exposing a connector to impact. It looks like this:

Start with the case:

Mount the Intel Edison on the Explorer Block, and plug the USB OTG cable into the USB port labelled “OTG”. Thread a regular USB cable through the hole in the case and conect it to the Explorer block, and put the Explorer block in the case.

Plug in the G4. It should fit in the slot tightly.

Next, we’re going to guide the cable to prevent the USB port on the Explorer block from getting damaged if the cable gets pulled on. (This risk is not theoretical; before I built this case, I broke the connector off one Explorer Block that way). This will also give limited protection from moisture. Start by placing two pieces of electrical tape over the hole, one on each side of the cable:

Add a horizontal piece of tape to guide the cable to the end, followed by a vertical piece to secure it in place. If you ever yank on the cable too hard, this will come off and absorb most of the force, protecting the rest of the rig from damage.

Next, tape the top. Be careful not to cover too much of the speakers, or alerts migth stop being audible. Finally, hook up the battery.

One limitation of this setup is that, because the battery is connected via USB instead of via the 3.7V power connector on the Explorer Block, the OpenAPS software can’t measure the battery’s charge level and warn you when it’s low. This is unavoidable, since if you power it via the 3.7V power connector instead, then it can’t communicate with the Dexcom G4, since USB OTG requires 5V power and it doesn’t have the circuitry to step 3.7V up to 5V. I’ve addressed this by adding a software feature which measures uptime, and delivers an alert when it’s been running for a duration which suggests the battery is probably low; since the rate of drain is fairly consistent, and since I’m swapping two identical batteries and always charging them fully before connecting them, this is good enough. This feature will probably land in OpenAPS 0.7, or (if you must) you can use my fork.

This is the second revision. The first was about 10% bigger on each axis than it needed to be, and had sharp corners. If someone wanted to make a third revision, the one thing that could be improved would be to carve out a compartment on the bottom for two AAA batteries, for use as pump spares.

People in my circle have been talking a lot about the Impossible Burger, which is a plant-based meat substitute designed to imitate the flavor of, and to compete with, ground beef hamburgers. Discussion so far has largely focused on how the taste compares to beef hamburgers. This is my attempt to compare the Impossible Burger to beef on nutrition. I’m not comparing price, since availability is still limited and the current price is unlikely to remain stable over time.

This post was difficult to write, and doesn’t reach any clear conclusions about which is better. This isn’t because I think they’re equivalent, or close to equivalent; rather, I think it’s probably the case that the Impossible Burger is either significantly more or less healthy than beef, but am left with considerable uncertainty about which way the difference goes. This is due to a combination of uncertainty about the Impossible Burger’s actual chemical composition (some important details aren’t disclosed) and uncertainty about how healthy beef is (the quality of research in this area is quite poor). Ultimately I was forced to settle with identifying the relevant differences, flagging many of them as subject to uncertainty and/or controversy, and leave it at that.

The Outside View

Both beef and the Impossible Burger fall in reference classes with possible health concerns. Beef is red meat; the Impossible Burger is a processed food. Red meat is correlated with bad health outcomes in observational studies (here’s a meta-analysis); unfortunately this is a study methodology with an extremely poor track record, and there isn’t really anything better to go on. Intervention studies exist, but none that are long-term enough to measure mortality, only indirect biomarkers, and they find no effect (meta-analysis).

Detailed Comparison

There are a number of differences between the Impossible Burger and beef, so to keep track of them all, here’s a nutrition-facts-panel-style table. More detail on each of the rows where there’s a difference below.

The Impossible Burger FAQ says that “bioavailable protein, iron, and fat content are comparable to conventional 80/20 ground beef”, but the fat content on the label matches more closely to 85/15 ground beef, so I’ll use that as the basis for comparison. (Beef is notated as percent-lean/percent-fat, and typically ranges from 93/7 to 70/30, with 80/20 or 85/15 most commonly used in hamburgers.)

Impossible Burger ingredients: Water, Textured Wheat Protein, Coconut Oil, Potato Protein, Natural Flavors, 2% or less of: Leghemoglobin (soy), Yeast Extract, Salt, Soy Protein Isolate, Konjac Gum, Xanthan Gum, Vitamin C, Thiamin (Vitamin B1), Zinc, Niacin, Vitamin B6, Riboflavin (Vitamin B2), and Vitamin B12.

Other information sources:

* Estimated based on ingredients list, but probably pretty accurate
** Estimated based on ingredients list, with a large error margin
† A difference that’s significant but where it’s unclear or controversial which is better

Fat

The fat in the Impossible Burger comes from coconut oil, which is mostly composed of medium-chain triglycerides (MCTs). MCTs are commonly used by people trying to induce ketosis – that is, to make their body use fat for short-term energy needs (as opposed to making cell membranes out of it, storing it in the liver for use over a longer time horizon, or storing it in fat cells for long-term storage). This is probably a good thing, but it’s unusual, and puts the Impossible Burger in a distinctly different nutritional niche than beef burgers, and may interact strangely with unusual metabolisms.

Beef contains both omega-3 and omega-6 fats, while the Impossible Burger contains neither. The standard line is that absolute quantity of omega-3 and omega-6 fats doesn’t matter, but ratio does. Unfortunately, there isn’t a clear answer to what the ratio or amounts are in typical ground beef; they depend on both the breed of cow and on what it ate (grass-fed beef leads to a more favorable omega-3 ratio), and the 3:6 ratios range from 1.8 to 13.6. The former is probably good, the latter is probably bad. Beef also contains vaccenic acid, which is a trans-fat. Trans fats as a category have been found to have detrimental health effects, but this was based on the distribution of trans fats that resulted from partial hydrogenation.

Overall, the Impossible Burger’s fat is probably better than beef fat.

Protein

The main dietary purpose of beef, as eaten in practice, is to provide protein. With regards to protein, the nutrition facts panels of ground beef and the Impossible Burger look pretty similar; beef has more, but only slightly (23.6g vs 20g per 85g serving). However, not all protein is created equal; proteins are made up of amino acids in varying proportions, and if something contains more of the amino acids you don’t need but lacks the amino acids that you do, that’s not ideal.

From the ingredients label, the protein in the Impossible Burger mainly comes from textured wheat protein and potato protein, with additional small amounts from leghemoglobin, yeast extract, and soy protein isolate. The proportions aren’t disclosed, nor is the overall amino acid profile. There’s a reasonably good justification for this; providing that information isn’t customary for processed foods, and their upstream ingredient suppliers don’t necessarily provide amino acid profile information either. Still, we can infer some things about the protein in the Impossible Burger based on the order of the ingredients list: at least 13g of textured wheat protein, between 2g and 7g of potato protein, at most 1.7g of leghemoglobin and yeast extract, and at most 430mg of soy protein isolate. The closest I could find to a public statement about the amino acid profile of the Impossible Burger is a tweet saying

The #impossibleburger contains all essential and non-essential amino acids. The amount of bioavailable protein is comparable to beef.

This isn’t entirely reassuring. Suppose they had taken a large amount of bioavailable but incomplete protein, and mixed in a small amount of complete protein. That would make the statement true, but it would be nutritionally poor. As it turns out, potato protein is a complete protein and wheat protein isn’t, so this might in fact be what happened. Depending what part of the plant you use, wheat has a Protein Digestibility Corrected Amino Acid Score somewhere between 0.25 and 0.53, as compared to 0.92 for beef. I checked out some producers of textured wheat protein, but they didn’t disclose their amino acid profile either.

Leghemoglobin

The component that gives beef much of its distinctive flavor is hemoglobin, which is a blood protein that transports oxygen in mammals. The Impossible Burger’s signature ingredient is leghemoglobin, a protein which provides a similar taste and color. Leghemoglobin was originally isolated from the roots of soybeans, and is produced using genetically modified yeast. Leghemoglobin is the Impossible Burger’s most controversial ingredient, largely because of publicity around Impossible Foods’ conversations with the FDA, which were publicized when an environmentalist organization, ETC Group, used a FOIA request to get them. This was then used as a platform to criticize FDA practices in general.

The FDA’s concerns, however, were specifically about allergenicity. With some searching, I was able to find one report of an allergic reaction possibly related to the Impossible Burger. I contacted the author of that post, and they reported that blood tests failed to identify any allergies, and the only time they’ve ever had an allergic reactions was the one time they tried the Impossible Burger. That is the only example of a possible allergic reaction to the Impossible Burger I could find; if there are others, they haven’t been written about publicly. And it’s somewhat speculative; it might’ve been a reaction to something else, since they (understandably) didn’t try it again. Overall, leghemoglobin doesn’t concern me very much; trying it doesn’t seem notably higher risk than trying any other new food processed food, and if there does turn out to be an allergenicity problem, I expect we’ll find out (with prevalence numbers) soon enough.

Miscellaneous

The Impossible Burger lists “natural flavors” on the ingredients above the 2% mark, meaning at least 2% of it consists of undisclosed ingredients from that category. Those ingredients could potentially be bad.

Beef contains creatine, which was found to have cognitive benefits when given as a supplement to vegetarians (but not to meat eaters).

The Impossible Burger contains 5g of carbohydrates (most likely as impurity in the textured wheat protein). If it’s on a bun, this is pretty small. It does make the Impossible Burger less suitable for low-carb/ketogenic diets, though.

Conclusion

Overall, the Impossible Burger seems good enough from a nutrition perspective to be competitive with beef. However, I have reservations about aspects of its composition that aren’t publicly known, particularly the amino acid profile and what might be lurking under the “natural flavors” heading. Limited availability means I won’t be eating this in the short term, but once it’s sold in grocery stores and reaches a price not too much higher than ground beef, I’ll probably start eating it.

Regarding what happened recently in Charlottesville –

Actually, first, I’m going to put this filler paragraph here, so that people looking for quick emotional engagement can go read something else. This post is about subtleties, and I would prefer that anyone who wants to eschew subtlety in favor of emotion unsubscribe from my wall.

That out of the way, Alex Fields, the driver of the car that ran into a crowd, is by all accounts a racist and a scumbag. I fully expect that he will be hung to dry, and I have no problem with this. However, given his prominence in the news cycle, I believe it’s important to be truthful and precise about exactly what he did. The story running in the news is that he committed premeditated mass-murder, by running into a crowd of counter-protesters with his car.

I make a habit of fact-checking news stories, both political and non-political, and calling out any errors I see. I decided to check this one. Was it really murder, or was it merely vehicular manslaughter? This is important, because falsely believing that other people have escalated the level of violence in a conflict can *cause* the level of violence to escalate. So, how strong is the case against Alex Fields?

I’ve looked at it from as many angles as I could find, and reconstructed the sequence of events as best I could. Using a stopwatch for time and Google Maps for distance, I (imprecisely) estimated the car’s speed. I sorted through YouTube’s terrible search to get non-duplicate videos. I watched the important moments of them frame-by-frame. Here’s the sequence of events:

1. Alex Fields is driving south towards a protest. There are pedestrians about 120ft ahead of him. His brake lights are on. The camera moves away from his car. and returns a second later.
2. In 3.1s the car moves 105ft (about 24mph).
3. Alex’s car is struck hard on the rear bumper by a stick or pole. Prior to this point he has not hit any pedestrians.
4. About 50ft ahead of Alex’s car is a white convertible; in between, there are pedestrians.
5. Alex’s car accelerates. Two pedestrians go over his hood, one is sandwiched between his car and the next, and several more on the sides are knocked over without being seriously hurt.
6. Alex’s car hits the convertible, and the convertible hits the car in front of it, a red minivan. The minivan goes into the intersection, hitting about ten pedestrians
7. Alex Fields shifts into reverse and starts backing up
8. About seven people charge the car, striking it with sticks, smashing the rear windshield and the front windshield on the passenger side
9. The car backs into two of the people who were attacking it, one of whom seems to end up seriously hurt

I think the most likely explanation is that, after the first strike on his car, he panicked. I am fairly certain that this will be his defense in court, and, unless evidence is found which is not yet public, I don’t think his guilt can be proven even to a preponderance of evidence standard.

Angles:
From behind: https://www.youtube.com/watch?v=zeB2ZaUSa48
From front: https://www.youtube.com/watch?v=r0guSatguSk
People hitting the car after it crashes: https://www.youtube.com/watch?v=9mnywjPPDtU
Drone view: https://www.youtube.com/watch?v=6QHXk2AB6kU

(Cross-posted on Facebook)

In 2000, the US presidential election between George Bush and Al Gore was swung by hacked voting machines. This fact can be verified by anyone, beyond a shadow of a doubt. In the close race between Bush and Gore, Volusia County reported a total of negative-16,022 votes for Al Gore^[1]. Upon investigation, the negative vote count disappeared and was replaced with a normal-looking one, but not before the number reached the press. At the time, this was unexplainable. It was not until 2005 that security researcher Harri Hursti demonstrated a technique for hacking voting machines that involves sneaking memory cards with negative vote totals on them into the counting process^[2]. The idea is that, by inserting a memory card with positive votes for one candidate and negative votes for another candidate, one can change the vote totals without messing up the turnout numbers, which would reveal the fraud. But if one is performing the Hursti hack, and messes up by putting the fake memory card into the process *in the wrong place*, then a county may accidentally report a negative vote total – because a memory card that was supposed to be used in a large precinct was used in a small precinct, without enough votes to steal. The machines used in Volusia County were in fact vulnerable to this technique, and this is what happened.

Because the margin in Florida was small, a “recount” was triggered; in reality, ballots were being viewed by humans for the first time. Bush and Gore argued before the Supreme Court, and the Supreme Court ruled 5-4 (along party lines) to stop the counting. Gore then conceded. It all happened too quickly for little things like election fraud to be noticed; the media narrative, rather than being about fraud, was about “hanging chads” and other innocent things.

Since then, forces within the Republican party have worked to promote a false narrative of vote fraud as something symmetric, that both parties do, by manufacturing false evidence of Democratic voter fraud. Like voting machine hacking, this becomes most obvious when there’s a mistake and the mistake becomes a scandal. In 2006, seven attorneys were fired mid-term from the department of justice^[3]. In 2008, the Inspector General determined^[4] that these firings were motivated by refusal to prosecute voter fraud cases against Democrats.
Over the past few weeks, Donald Trump has been loudly warning that democrats would engage in election fraud to give the election to Hillary Clinton. This is a possible thing, but so is the reverse. Fortunately, there’s a standard strategy for determining whether election fraud took place, and which direct it went in: exit polls.
I don’t have access to exit poll data for today’s election. Neither do you, and neither do most of the news outlets that’re reporting the results. But Nate Silver has this to say about them on his blog:
“One reason people find Trump’s competitive margins across a wide range of swing states so surprising is because exit polls showed Clinton beating her pre-election polls in most states, instead of underperforming them.”^[5]

At the time I’m writing this, FiveThirtyEight says (on its sidebar) that Clinton and Trump are equally likely to win, based on states that have been officially called; other sources are strongly favoring Trump’s chances, based on preliminary counts.

I roll to disbelieve. Maybe Trump really did get more votes than expected, and more votes than indicated by the exit polls Silver was referring to. Or maybe he didn’t. One thing’s for certain, though: the American people will hear a definitive result before any of this is investigated.

(Cross-posted on Facebook)

Last Wednesday, Google announced that AlphaGo was not powered by GPUs as everyone thought, but by Google’s own custom ASICs (application-specific integrated circuits), which they are calling “tensor processing units” (TPUs).

We’ve been running TPUs inside our data centers for more than a year, and have found them to deliver an order of magnitude better-optimized performance per watt for machine learning.
…
TPU is tailored to machine learning applications, allowing the chip to be more tolerant of reduced computational precision, which means it requires fewer transistors per operation.

So, what does this mean for AGI timelines, and how does the existence of TPUs affect the outcome when AGI does come into existence?

The development of TPUs accelerated the timeline of AGI development. This is fairly straightforward; researchers can do more experiments with more computing power, and algorithms that stretched past the limits of available computing power before became possible.

If your estimate of when there will be human-comparable or superintelligent AGI was based on the very high rate of progress in the past year, then this should make you expect AGI to arrive later, because it explains some of that progress with a one-time gain that can’t be repeated. If your timeline estimate was based on extrapolating Moore’s Law or the rate of progress excluding the past year, then this should make you expect AGI to arrive sooner.

Some people model AGI development as a race between capability and control, and want us to know more about how to control AGIs before they’re created. Under a model of differential technological development, the creation of TPUs could be bad if it accelerates progress in AI capability more than it accelerates progress in AI safety. I have mixed feelings about differential technological development as applied to AGI; while the safety/control research has a long way to go, humanity faces a lot of serious problems which AGI could solve. In this particular case, however, I think the differential technological advancement model is wrong in an interesting way.

Take the perspective of a few years ago, before Google invested in developing ASICs. Switching from GPUs to more specialized processors looks pretty inevitable; it’s a question of when it will happen, not whether it will. Whenever the transition happens, it creates a discontinuous jump in capability; Google’s announcement calls it “roughly equivalent to fast-forwarding technology about seven years into the future”. This is slight hyperbole, but if you take it at face value, it raises an interesting question: which seven years do you want to fast-forward over? Suppose the transition were delayed for a very long time, until AGI of near-human or greater-than-human intelligence was created or was about to be created. Under those circumstances, introducing specialized processors into the mix would be much riskier than it is now. A discontinuous increase in computational power could mean that AGI capability skips discontinuously over the region that contains the best opportunities to study an AGI and work on its safety.

In diagram form:

I don’t know whether this is what Google was thinking when they decided to invest in TPUs. (It probably wasn’t; gaining a competitive advantage is reason enough). But it does seem extremely important.

There are a few smaller strategic considerations that also point in the direction of TPUs being a good idea. GPU drivers are extremely complicated, and rumor has it that the code bases of both major GPU
manufacturers are quite messy; starting from scratch in a context that doesn’t have to deal with games and legacy code can greatly improve reliability. When AGIs first come into existence, if they run on specialized hardware then the developers won’t be able to increase its power as rapidly by renting more computers because availability of the specialized hardware will be more limited. Similarly, an AGI acting autonomously won’t be able to increase its power that way either. Datacenters full of AI-specific chips make monitoring easier by concentrating AI development into predictable locations.

Overall, I’d say Google’s TPUs are a very positive development from a safety standpoint.

Of course, there’s still the question of how the heck they actually are, beyond the fact that they’re specialized processors that train neural nets quickly. In all likelihood, many of the gains come from tricks they haven’t talked about publicly, but we can make some reasonable inferences from what they have said.

Training a neural net involves doing a lot of arithmetic with a very regular structure, like multiplying large matrices and tensors together. Algorithms for training neural nets parallelize extremely well; if you double the amount of processors working on a neural net, you can finish the same task in half the time, or make your neural net bigger. Prior to 2008 or so, machine learning was mostly done on general-purpose CPUs — ie, Intel and AMD’s x86 and x86_64 chips. Around 2008, GPUs started becoming less specific to graphics and more general purpose, and today nearly all machine learning is done with “general-purpose GPU” (GPGPU). GPUs can perform operations like tensor multiplication more than an order of magnitude faster. Why’s that? Here’s a picture of an AMD Bulldozer CPU which illustrates the problem CPUs have. This is a four-core x86_64 CPU from late 2011.

Highlighted in red, I’ve marked the floating point unit, which is the only part of the CPU that’s doing actual arithmetic when you use it to train a neural net. It is very small. This is typical of modern CPU architectures; the vast majority of the silicon and the power is spent dealing with control flow, instruction decoding and scheduling, and the memory hierarchy. If we could somehow get rid of that overhead, we could fill the whole chip with floating-point units.

This is exactly what a GPU is. GPUs only work on computations with highly regular structure; they can’t handle branches or other control flow, they have comparatively simple instruction sets (and hide that instruction set behind a driver so it doesn’t have to be backwards compatibility), and they have predictable memory-access patterns to reduce the need for cache. They spend most of their energy and chip-area on arithmetic units that take in very wide vectors of numbers, and operate on all of them at once.

But GPUs still retain a lot of computational flexibility that training a neural net doesn’t need. In particular, they work on numbers with varying numbers of digits, which requires duplicating a lot of the arithmetic circuitry. While Google has published very little about their TPUs, one thing they did mention is reduced computational precision.

As a point of comparison, take Nvidia’s most recent GPU architecture, Pascal.

Each SM [streaming multiprocessor] in GP100 features 32 double precision (FP64) CUDA Cores, which is one-half the number of FP32 single precision CUDA Cores.
…
Using FP16 computation improves performance up to 2x compared to FP32 arithmetic, and similarly FP16 data transfers take less time than FP32 or FP64 transfers.

So a significant fraction of Nvidia’s GPUs are FP64 cores, which are useless for deep learning. When it does FP16 operations, it uses an FP32 core in a special mode, which is almost certainly less efficient than using two purpose-built FP16 cores. A TPU can also omit hardware for unused operations like trigonometric functions and probably, for that matter, division. Does this add up to a full order of magnitude? I’m not really sure. But I’d love to see Google publish more details of their TPUs, so that the whole AI research community can make the same switch.

When people organize to commit a crime, the legal term for this is a conspiracy. For example, if five people get together and plan a bank robbery with the intent to execute their plan, then each of them has committed the crime of “conspiracy to commit robbery”.

If you suspect that a crime has taken place, and it’s the sort of crime that would’ve involved multiple people, then this is a “conspiracy theory”, particularly if those people would’ve had to be powerful and well-connected. Until recently, everyone seemed to agree that conspiracy theories were exclusively the creations of idiots and schizophrenics. One of the more common, well known conspiracy theories is literally that “the CIA is causing my command hallucinations”. So if someone were to claim, for example, that the 2000 election was swung by hacked voting machines in Volusia County, or that Michael Hastings was murdered by one of the top-ranking military officials he singled out for criticism in Rolling Stone? Then the anticipation is that this would be followed by some incoherent rambling about aliens, the social expectation is that this person is to be shunned, and the practical effect is that those things are hard to say, hard to hear, and unlikely to be taken seriously.

Somehow, somewhere along the line, accusing governments and the powerful of crimes became associated with mental illness, disrespectability and raging idiocy. And when you pause and notice that, it is *incredibly creepy*. It’s certainly possible to imagine how such a meme could arise naturally, but it’s also a little too perfect — like it’s the radioactive waste of a propaganda war fought generations back.

As it turns out, the last few generations did have propaganda wars, and we know what they were: the counterculture, the civil rights movement, McCarthyism. Each side of each memetic conflict left its mark on our culture, and some of those marks are things we’d better scrub away. So now I’m wondering: what were they?

(Cross-posted to Facebook)