techdragon

Typically yes... Alternative mechanisms include using solar panels to increase drag, and some companies are experimenting with devices that interact with the earths magnetic field to produce electromagnetic drag…

But reserving fuel for “decommissioning operations” is standard practice for satellite operators and the space industry in general.

Touching stuff...

It could be middle age tainting my perspective... but sometimes I feel this sort of... not visceral, but very close and undeniable... experience with death is easier for people who are young.

Death isn't real for most people at 17, no matter how much they have seen, many see the death of a grandparent, or two, or perhaps even four, before reaching 17, but most still enter their late teens, feeling invincible and like death is a whole lifetime away... because on a sort of undeniable personal narrative level is is... it is a lifetime away... and things that feel a lifetime away are much easier to think about than things that feel like they get closer every year. In middle age you know your to some statistical average "about half way there"... and that's a very different place from which to begin having this kind of life experience.

As a long term DataDog customer… no it’s beyond that.

DataDog have an extractive pricing model. It’s undergone a number of changes over the years that separated out features previously under the host price into their own products with separate monthly costs. They put limits on the numbers of unique containers you can have per host before charging container fees, so better hope you don’t have too many crash-restart loops or. Theres the per unique host per billing period pricing model that makes it impossible to scale your own infrastructure up and down to save on hardware costs, one personal case I can attest to was case the changes would halve the AWS bill but increase the DataDog bill by more than 10 times, a thousand percent… and you would think surely you can talk to the sales team if your use case fits extremely poorly with the pricing model…. Hahaha no… unless your a whale they give zero fucks about giving you any flexibility on price.

The issue I’ve had with the BBB is that they always seemed to be a premium cost item. Why am I paying more for something that gives me less compute than a most of the RPi boards. I have one and when I got it it cost me something over $150 AUD, I don’t remember the price exactly but I remember how ripped off I felt, at the time the premium RPi 3 was under $100 AUD

I’d the BBB cost me ~$50 AUD I’d be much more keen on them… for all the well outlined reasons. It it just seems to always cost more than I’m prepared to pay for that hardware.

It would be very funny if Magic Mushrooms became schedule 2 before cannabis does … feels very emblematic of how much social/class warfare is co-opted into by vested interests and the whole racism by proxy elements of the war on drugs…

This is Crack-Babies 2: Fentanyl Boogaloo

“Six individuals born after pregnancies complicated by multiple drug exposures, including fentanyl shared physical and facial features suggestive of a novel syndrome. “

So in other words, it could be fentanyl…

Or possibly the other drugs…

Or possibly the combination of drug exposure…

Or a contraindicated drug reaction between the non-prescription drugs…

Or and just stay with me now… possibly because of some unknown chemical used to “cut” the non-prescription drugs they were talking… because in case anyone hasn’t picked up on it… this one of the polite clinical ways of saying without calling them criminals… that these mothers acquired one or more drugs illegally for their own purposes be they pain relief or recreational getting high.

I’ll say it again, this is some sloppy “crack baby” shit, and they should be extremely careful about what they are saying due to current fentanyl related psychogenic hysteria and not writing up their research in such a way that it unethically fans the flames of inappropriate public health outcomes when they have a cherry picked sample size of fucking 10 and nothing more certain than self reported drug use histories.

My very biased advice is leverage your public work, and just go freelance. You have to develop some other skills like networking if you want to avoid the meat market freelancing job sites… oh and don’t do websites unless there’s something other than the website involved, or they agree up front to a minimum hour per week maintenance deal (like they will pay 1 billable hour per week regardless of if they talk to you at all, to cover your time spent keeping on top of patches and security) there’s so little value add in website building that you quickly find people constantly using up your time asking for trivial changes that without a maintenance deal will start to make the customer resent paying for them and it’s not worth the headaches…

There’s a lot of good money out there for people with experience who are willing to help small businesses who have sufficient IT skills to have reached the point where they want to automate and improve their workflows and management processes, who don’t feel the off the shelf products like excel, Airtable, etc, are a good fit or do the job they need, and who are willing to either pay either hourly for the work or enough per contact that it’s worth the risk you’ve estimated wrong…

This is obviously much harder in the USA due to the symbiotic (being generous about the positive connotations) nature of employment and healthcare.

I just want to call out the how nice it is to have a global view of all the keyboard shortcuts like this. Including for applications you might not have open. It makes it a million times easier to fix up situations where two applications want to use the same keyboard shortcut and that causes some kind of issue… which is usually not a problem but can be a major pain when one or more of the conflicting shortcuts are global shortcuts that don’t require the application to have focus or be in the foreground.

Being able to jump into this settings pane and find the issue then simply remap one … infinitely easier than fixing this kind of thing on Windows… where last time I tried literally none of the “list keyboard shortcuts from all open applications” tools I could find worked at all on Windows 10.

Yeah, since lead accumulates in the body, I can imagine that a lifetime of exposure from significant sources like water pipes and all the other ways Roman’s were using lead, will just slowly add up until it reaches these sorts of levels, one microgram at a time until you’ve got like two tenths of a kilo of lead in you and a few organs are starting to struggle.

You’ve got a solid point about how I was oversimplifying, but it’s not an “inaccurate picture” it’s just not providing as much detail as is needed to accurately predict real world effects… which unless you’re actually working with radiation environments that where the radiation flux is high enough to have noticeable impact before the hardware becomes obsolete or otherwise gets replaced, it’s not a bad analogy.

I’ve had to talk to plenty of people with various levels of understanding of this, including smart hardware and FPGA types, a few radiation physics people as well, it’s not a well established “discipline” outside of some extremely small teams of engineers responsible for the sort of small batch hardened mil-spec chip fabrication runs that doesn’t really get a lot of public documentation, because while not classified, it’s extremely commercially sensitive since and the things the chips go in are usually classified to some level so there’s not a lot to work with other than some stuff that comes out of detector physics at CERN and a trickle of research papers that have come out over the years from NASA, some old DITC papers, the odd helpfully detailed SBIR grant, and some bits here and there in the occasional physics or semiconductor engineering research papers where someone gets into interesting theoretical modelling or analysis of flown hardware or something like that… it’s a niche within a niche within a niche … and the reason I know anything at all beyond being a space nerd for most of my life that was also a computer geek… is that I decided to create a startup doing space robotics, and it should be no surprise that modern robots worth the cost of putting them in space need a respectable amount of integrated circuits and all sorts of other electronics.

I think your making it seem more complicated than it really is beyond the niche nature of knowledge on the topic. It’s the typical aerospace trade triangle: sophistication vs size and weight vs cost. Except in “miniature” and having implications on software design and development. You can shield the hell out of small sensitive components but that’s going to take up size and add mass, you can get military spec parts that take up barely more room than the normal ones do but that’s going to cost a lot, you can go old school and use simpler circuit designs or larger components but this may save some money on components but increasing the size or weight which will increase the cost to get it into space…

I did not… it seemed wildly high to me too.

But I checked and got second sources that matched close. So I rounded down from 5670mg/kg to a more easily to do mental math with 5500mg/kg.

It turns out the amount of lead required to kill you directly from lead poisoning is… quite a lot! But that’s kill you… your probably going to have some other lead poisoning related health problems well before that.

Edit: Re googling one of my references.

“LD50 for lead 4665 mg/kg of bodyweight in males and 5610 mg/kg of body weight in females.” from https://www.aatbio.com/resources/faq-frequently-asked-questi...

Having an expert (or other more specialised software) optimise an otherwise satisfactory 3d model is going to be vastly more simple a process than what can be a painstaking process of getting the “original vision” for a shape into a digital form.

I’m better with CAD style workflows than I am with Freeform modelling, but I’m also not an experienced enough user of any particular CAD program that I don’t spend ages frustratingly trying to understand where to go to edit the correct part of the chain of operations that built up a particular fillet, chamfer, or sweep, every program is a little different… just different enough to throw me off since I don’t use them often enough to really get over that learning threshold, for instance I’ve used Onshape one time in the last 12 months… spent an hour trying to work something out… and then 5 minutes actually making my edit…

If I could have opened up the design in 2d blueprints form, and put in some kind of multi modal query to the effect of “these are my blueprints, can you convert them to (CAD program of choice), and while you are converting them, increase cross section CS1 vertically to from 8mm to 10mm keeping everything else fixed” … would save me a lot of time making adjustments to peoples 3d models for printing. It can be really annoying to fuck around as much as is sometimes necessary due to individual cad workflow preferences just to make conceptually simple edits like “this hanging pot, but stretched 50% taller, and with all the holes/cutouts kept the same size please”

The ML models might not be able to do a perfect curve for a gorgeous arcing buttress or a complex bolt hole pattern for attaching multiple elements to a primary support structure… but it should be able to do a lot of the sorts of work that people use OpenSCAD for.

While it’s similar to the kinds of biological damage that radiation can cause, because the radiation induced damage in DNA does work similarly, the energy from a radiation “hit” breaks bonds in the DNA molecule and causes knock on effects just like in the silicon crystal structures and metalic wires…

The better biological analogy for the slow degradation of the integrated circuits due to radiation exposure would probably be heavy metal poisoning because the damage doesn’t typically spread or multiply like DNA damage and/or cancer does. Heavy metal poisoning “accumulates” it’s a slow buildup of something not meant to be there, which causes increasing problems with biological systems, until something is s disrupted by the heavy metal concentration giving a progression of symptoms until enough has built up and you get fatal symptoms.

Edit: Simple example/comparison

Everyone can (and we sadly have plenty of evidence for this) tolerate a fair bit of lead with an LD50 of 4500 to 5500 mg/kg or and a little bit of mercury with various mercury salts having an LD50 from 6-200 mg/kg, (LD50 is the abbreviation for "lethal dose, 50%", The LD50 for a substance is the dose required to kill half the members of a population.). Two heavy metals, two different levels of "exposure risks".

Similarly the metal & doped semiconductors of an integrated circuit have have different sensitives to different kinds of radiation damage. Metal interconnects will tolerate radiation differently than the silicon which is different again to the insulation substrate...

A Heavy ion, like a stray iron nucleus, (https://en.wikipedia.org/wiki/HZE_ion) is going to just dump a lot of energy and screw up a chunk of the atoms in the area around of whatever and wherever it hits, its like the mercury salts, you cant take a lot of hits like that, and fortunately they are statistically rare.

Higher energy protons, like cosmic ray protons, can cause proton induced transmutation which will slowly turn the atoms you want into the atoms you don't want, and can cause the silicon transistors themselves to change similar to how we can (and sometimes do) use neutrons to dope semiconductors in the first place (https://en.wikipedia.org/wiki/Doping_(semiconductor)#Neutron...)

Lower energy protons, and high energy electrons and positrons, can also lead to a slow buildup of electrical charges which can cause dielectric breakdowns and other unwanted effects, which can depending on what is involved and how it happens the cause of intermittent or permanent effects.

Three different kinds of radiation, three different exposure disks. They are each like different "poisons", and the total exposure, the total absorbed dose of each of them accumulates over time... until one or more of them cause something critical to happen and render the hardware inoperable.

For certain niche applications… this is actually a very good thing.

Case in point. Radiation hardness, the ability of a chip to resist the long term effects of radiation damage is highly related to the cross sectional area of the wires. The shiny efficient circuits we make now with their tiny single digit nanometer surface features… are getting to the point where we’re only working with something a few dozen atoms wide. Which means that when high enough energy radiation slams into that part of a circuit, even if all it does is knock a single atom out place or change its atomic structure by neutron or proton capture or splitting it into two or more new atoms… regardless of the result your damaging an appreciable portion of the circuit and It won’t be able to take much before the cumulative effects of exposure to all the radiation degrades the integrated circuit traces and causes eventual failure. A lot of space grade hardware gets around this with having secondary backup hardware or using FPGAs so damaged silicon can be routed around over time. But this gets expensive to use for everything…

Having a way to build basic components like power management watchdogs and other critical circuitry that doesn’t require a lot of “smarts” but does need to be reliable… definitely could benefit from this sort of technology.

I had this “should be 30 minutes” but is really “never”… after getting an AirPod replacement in August this year.

In my case it was actually a pretty easy to fix once I knew what it was. You have to do a little fiddling to check this since it involves a couple reset/pairing cycles with just a single AirPod in the case … but if you check I bet the replacement has an older firmware since it’s been sitting in the stores spare parts room.

So you trigger the firmware update with just one AirPod in the case (which was pretty easy and I could just follow the Apple information) and then once the firmware is the same then yeah they will both do the automatic pairing like the Genius Bar tech expected them to… the issue might be related to a recent firmware changing something and the stores not having updated process to ensure the firmware is correct because apple are probably going to prefer this having a software fix not a fix that takes up more human store employees time.

This fuzziness is slowly getting worse in my opinion. I notice it seems like Google had gotten more and more willing to assume unrelated words/concepts are sufficiently interchangeable that it can happily return both in a search query for either … and I’ll be honest here… single behaviour is the number one reason I’m on the edge of leaving google search forever… doing a takeout export of my gmail and fucking the hell off to only using iCloud for personal and Microsoft 365 for work and professional…

Google’s continued erosion of their core user facing product “search” (the real core product is advertising but that’s not the majority of people interacting with google are interacting with google for… and an argument could be made that googles only real product customers care about anymore is YouTube but the quality of that experience fluctuates wildly depending on how stupid they are being any particular day due to asinine policies and abusive relationships where they seem to desperately want to destroy the goodwill of the creators that upload content like it’s some kind of fetish and they just have to know the creators hate them otherwise the job of working at YouTube isn’t satisfying…) is indicative of a complete failure to care about the core competency of the company… and companies that fail to care about their core competency are rotting hulks doomed to die…

I really wish these chips weren’t so IO constrained.

I routinely check on the state of things for building a tiny NAS for some hardware redundancy reasons and despite significant improvements in CPU and GPU a power across the ARM ecosystem… IO speeds for storage still feel stuck in the distant past.

But Amazon don’t make this obvious or as easy as writing a review. This is an incentive mismatch at work that just happens to work in favour of Amazon and their shitty sellers… cough enshitification cough

If they actually wanted to manage the issue the option to report the product you received to CS would be right there as close to next to the leave a review button as possible.

Except outside America where it’s metric… also as someone who’s gone hunting for commercial space before, it’s kinda only 97% true that it’s always per year… there’s an irritating countercurrent of people that try and trick you into thinking something overpriced is a bargain by quoting the inflated prices at 1/12th the cost because that’s the monthly not the yearly, and people who just deliberately want to game various “by lowest price” sorting mechanisms on property listings websites.

Started writing a comment then noticed how much of the article is just quotes, so I’ll summarise instead of directly quoting and recommend reading it for additional context.

These kids are exactly the sort of overachievers and desperate to climb the ladder types you are probably thinking off… They didn’t need to drink the LinkedIn corporate-positivist cool aid, they were born in it, raised by it.

They are fully and un-ironically engaging in the performative positivism that makes LinkedIn intolerable to most adults with a greater degree of life (and adult social interaction) experiences… and because of how bad other social platforms have become some of them seem to like the little happy bubble that LinkedIn creates. The fact LinkedIn is just a different kind of toxic lie filled cesspool is not something they have realised yet because, to borrow an apt phrase I once heard “the naïveté of youth blinds you to the truth”, and they haven’t yet developed sufficient cynicism to see how shallow, meaningless and dishonest the majority of LinkedIn content is.

There’s a quote at the end from someone who sees the problem for what it is, the fact kids are now getting pressured into the business world before even getting into College or postgraduate work. But the standout quote is from their exemplar teen that likes linkedin where they quote one of his humblebrags, and it’s at that point it becomes obvious to me that that kid is already a walking talking business adult who has had their childhood destroyed before they even left college.

It’s completely impossible to get users to pay for it, the users that would pay can’t possibly justify our investment in this amazing technology… /sarcasm

Its probably not the only component... but a fun fact is that the Earth's rotation is affected by the average equatorial ocean temperatures, (anyone seen the ocean temperature charts this year... something something ships and sulfur dioxide accidental geoengineering that we stopped doing) warm ocean water displaces slightly more volumetrically and thus by way of its lower density, affects the spinning mass balance of the Earth like a ice skater curling in their fingers rather than their whole arms while spinning like a top, the effect is there... just so very subtle.

As mentioned in a reply to a sibling comment. GPS time is (to a human relevant usable level of accuracy) TAI minus 19 seconds. The issue is that you can't really "run TAI" on something, most firmware isn't designed to keep track of time this way and so expects to be working with UTC, the differences are subtle and basically boil down to "has no leap seconds and requires some math to accommodate this fact" and "expects leap seconds to exist, and be handled by humans and synchronization tools"... the latter which is fundamentally assuming that time is non-monotonic is at odds with the way that hardware clocks work in silicon and circuity.

So while you can design software to record things in TAI, and if you build your own hardware and firmware use TAI on that hardware and firmware... at a normal day to day level you can't really switch... but you can join the slowly growing collection of educated and aware people that appreciate the UTC/TAI mismatch in our software systems and push in the correct direction. The arguments about freezing UTC and not adding any more leap seconds to it are on the face of it, just stupid, we have GPS time which is fixed to TAI, we have TAI itself... we don't need to fundamentally change UTC just because we fucked up all the computers but just don't want to fix them properly. This is one time where counter to the Simpson's meme, the children are in fact wrong. UTC is harder for the computers to use, and that's because UTC was the wrong choice, so the answer is to stop having the damn computers use UTC as the internal time representation and switch to having them use an appropriate internal time representation.

I mean look... we can start getting really fun with it and get into the metrology of it all, but I'm having some fun and trying to spread the good word here.

Early programmers totally fucked up using UTC instead of TAI which as a monotonic count of seconds is a far more accurate representation of how computers keep track of time via continuous oscillation counters than any approximation of UTC until we developed widespread network synchronized NTP based systems and began to use UTC via these, and even then its not a really good representation because all these systems want to work with clean contiguous "one second is one second" time keeping, which is what monotonic time scales are, they inherently are this. So the choice of using a non-monotonic time scale on top of hardware that keeps time monotonically was flawed from the start.

We can get into the Sidereal time, Ephemeris time, the International Earth Rotation and Reference Systems Service, UT, TT, ERA, DUT1, MTC (Marian Co-ordinated Time), and what the difference is between an instance and an interval that represents a specific moment in time with an any temporal uncertainty attached... I do freely admit I'm a total unrepentant geek about this sort of thing, the only reason I don't own an atomic clock is because I just cant justify the few grand to buy one (cheapest I've seen is $1.5kAUD for an DIY module from Microsemi) when I know how precise my GPS based Stratum 0 embedded Linux NTP server is... Heck the IERS Bulletin emails are some of the only automated emails I look forward to and open up with a little glee, finding out what's up with the literal rotation of the planet, neatly dropped in my inbox!

But as nerds who already know a bunch of metrology/timekeeping stuff (which I am assuming from your comment) we could launch ourselves down the rabbit hole, and hope the readers can keep up, or as I've tried to with my comments here (particularly with the slightly whimsical choice of wordplay)... Try and give causal nerds who have only just tripped over the very idea that there is a better timescale that computers should be using instead of UTC some idea where to begin reading, and to wash away some of the false truths (like that GPS has leap seconds) and while technically TAI might not be easy for computers to use due to the retroactively corrected nature of it... at the end of the day all timekeeping is subject to measurement error, approximation, and correction either due to or in correlation with external sources, and thus because we only periodically correct computers to UTC mostly (on average) between 1hr to 1day via NTP depending on OS and preferences, correcting a monotonic system clock to the daily correlated and corrected TAI would be infinitely better due to eliminating all the leap second complaints, than repeatedly correcting the system clock to UTC and continuing to have the leap second problem, because UTC has a place in the timekeeping world a place that its miss-use by computers has now put in jeopardy (the calls driven by tech companies to fix UTC and have no more leap seconds).

Prior to the invention of NTP, system administrators would have just corrected it to TAI manually, and done so just as accurately as we would have corrected it to UTC, so arguments like "it would be harder to use TAI than UTC" have honestly never been true, its really just a problem of awareness and now that we're in a world where system firmware is broadly speaking designed to keep time via UTC and thus when you interrogate "system time" you get UTC, everything built up from this layer is "wrong" so its a bootstrapping problem on the level of switching to IPv6, yes its technically better, but none of the hardware supports it and so why bother when I can just keep hacking away at UTC/IPv4... except there's no UTC scarcity to drive any sort of adoption, so all that can be done is to proselytize the benefits of TAI to the best of ones ability and hope that eventually it makes a difference.

It is indeed “International Atomic Time” in English, for what it’s worth I chose to use the French version of the acronym because using the English text requires more explanation why the acronym for “International Atomic Time” is not IAT but TAI, than just using the French name which matches the acronym and letting the curious google some more about the topic.

Also for bonus points the curious should read about UTC and how no one is precisely sure why that’s the version we got, spoilers: The English acronym and the French acronym both aren’t in the correct order for us to use the letters UTC, yet the acronym is UTC and no one is sure exactly why, like it may have been a compromise hashed out over drinks at some hotel bar between metrologists at a conference and written into the text that way without comment… it’s lost to history (at least the history I’ve been able to find when I ran down this rabbit hole)

Not really disagreeing just expanding on the conversation…

I’m pretty happy with computers storing time as a “weird number that we add another number to to get the human time”… because we are already usually storing it as second/millisecond/etc offset from an arbitrary fixed point in history… we’re talking about going from “current integer (prefix)seconds value -> (some math involving leap year formulas and fiddly stuff not just x/365/24/etc) -> human display value” adding one single subtraction step to the some math part that that already has a bunch of other fast assembly arithmetic calculations involved in it.

As for divergence. We have two kinds of timekeeping. “Calendrical timekeeping” which will always get fixed to be close enough to the solar year that no one has a problem with it, and “Chronological timekeeping” which is basically the science of keeping a track of how many seconds have passed between arbitrary moments in time. All Calendrical timekeeping have for decades now been built on top of TAI and the science of Chronological timekeeping where metrologists have kept fastidious watch over atomic clocks and matched the beating of their resonating frequencies to the astronomical record keeping to allow ever more precise astronomy and the byproduct of this effort have been precise records of how long in scientifically precise seconds each day has been, and from these records we know we have to occasionally “add” a second if we want the “human time” to stay within 1 second of the astronomical time. Consequently we can simply decide we are happy with more or less drift. One second is small enough we can kinda just deal with it, but imagine having a leap minute every hundred years? Coordinate that mess, or how about a leap hour once a millennium? It’s trivial for people we don’t really time anything as humans faster than a second, so if we skip one it’s pretty much ok… it’s categorically on an issue for computers… which should never have used UTC and should have always used TAI because TAI doesn’t skip a second and so the fix would be “a fixed lookup table or a single if branch based on the date and the number of integers to add when doing the integer representation to human representation mathematics before updating the clock on your screen.

GPS never had leap seconds. Your hell is imaginary. It is a myth and a lie perpetrated by those that never looked beyond the convenient lie to children that is UTC and tried to understand the true nature of timekeeping.

GPS Time signal = TAI minus 19 seconds

https://aviation.stackexchange.com/a/90844

You have been lied to… your true enemies are the ones that continue to build monotonic systems on top of non monotonic time scales… Join us in the Temps Atomic International! TAI is the one true timescale.

Computers should be using TAI and converting to UTC when humans want to see a local human time. My sincere belief is this was a bigger mistake than null, bigger than all the Y2K “we stored the year as 2 digits” stuff… The fact that from the BIOS/system Firmware up “we” decided to use a fuzzy time scale that changes on an unpredictable but just infrequent enough almost everyone can ignore it even exists except for when it changes and they remember it exists… genuinely a poor choice that I suspect is in large part due to a lot of people not knowing about TAI

Love clicking a GitHub icon to go read the source code and learn more… and being taken to a google cloud storage bucket where there’s four files, a readme pdf that’s basically just links to several different google sheets documenting data that went into this, two ipython notebooks, and a pre trained model file in a big zip file which I can already tell will be the usual zipped pickle of a massive tensor flow model that deeply frustrating to work introspect and examine the code behind…

This feels like the GAN/capsule networks papers and “releases” I’d look at when I first tried to learn more about machine learning 4-5 years ago, back in then “Middle Ages” of ML, where the paucity of sources enriched the careers of Machine Learning experts who had no incentive to make their work really accessible beyond puff piece show off papers and a doi/citation to pad their h-rank or a link to the startup that funded it for their flimsy product concept that based on historical evidence was over 90% likely to fail to go beyond concept or prototype demo stage.

Edit! I might sound a bit bitter or annoyed but it’s because I’ve been getting into local ML for privacy and personal productivity and really trying to understand the capabilities and the hardware required for those capabilities… I’ve been particularly interested in potential robotics applications and I’m following this particular branch of ML intersecting Robotics quite closely as it affects the space robotics startup I founded, and it’s undoubtedly interesting work… I’m just deeply disappointed at how much of a throwback to a “worse” time this particular “publication” feels like. It’s not light on details of how they actually built it, the code feels like a throw it over the wall afterthought, and the distribution mechanism has had no effort put into making the material any more accessible, if the ipynb files were in a GitHub repo at least the notebook would be rendered so I could just read it in my browser without downloading and opening up the notebook in a code editor… it feels like there’s a code repo somewhere with a script that dumps these files into this bucket and that’s the repo where they run everything in Google colab and the strong association to Google’s DeepMind team makes me think that once I do open this up I’m going to find something that’s welded solid to googles custom TPUs and is basically useless to me… hence the mild vitriol because the entire publication gives me a bad taste in my mouth despite being extremely interested in the topic and willing to put work into understanding the current state of the art.

It did try to handle it, by coming to a complete stop as fast as possible and avoiding further movement. I definitely think it should be smart enough to not stop “on top of an obstacle” under the car… but as far as everything before it stopping on the pedestrians leg… seemed like it would be hard to get a human driver to do better given the likely lack of a lot of “hit and run happens right in front of me” training data… zebras vs horses and all that. I expect one they will definitely add more “accident close by” data to the training data… and two they will get the damn thing not to stop on top of an unknown obstacle.