HN user
Between METI, Toyota & others, they appear to have spent > $40B over multiple decades.
It's a bit silly to downplay their investment based on the sales numbers of Toyota Mirai.
Some geopolitical context:
https://en.wikipedia.org/wiki/2010_Senkaku_boat_collision_in...
https://www.rusi.org/explore-our-research/publications/comme...
Japan was one of the first countries to be hit with rare-earth export-restrictions by China - going back to 2010. It seems that a lot of policy came out from this unpleasant shock, incl. the decision by Toyota to focus on developing FCEVs which would be less dependent on Chinese supply-chains. Ironically, the resulting vacuum may have actually led to Chinese/American companies gaining market share in the BEV space.
Still, given how things are going, FCEVs (and Japan with it) might actually end-up having the last laugh.
Au contraire - LLMs are quite bad at large scale pattern fidelity. They'll even forget key details and constraints unless told over and over again. That's why AI-written code has the quality of a patch-on-patch-on-patch.
This seems to be a redux of the 1x play (which was panned across the internet after the WSJ review).
https://www.youtube.com/watch?v=f3c4mQty_so
The hardware for these machines have been capable of household tasks for atleast 2 decades.
Here's PR1 cleaning up a room,
https://www.youtube.com/watch?v=o7JH3UWO6I0
The issue is not even teleop as a product. The issue none of these companies talk about is one of state-reset. Even a teleop-ed robot comes nowhere near the dexterity of a human - as the Joanna Stern review of 1x shows: 10 mins to load a dishwasher, 5 mins to get a glass of water, body coming the way of a fridge door, irrecoverable breakdowns every 30 mins...
Consider what happens if it drops some glassware or spills liquids on a carpet in addition (or worse does something stupid with the kitchen appliances). The teleop guy in Phillipines or India can't hop on a plane to fix this.
This 'environment reset' problem is at the core of RL - there are no solutions for this yet, only workarounds.
Yup, it's basically the ROS of computer-vision.
The article has the "rough" idea correct - how could it not be- but is quite stupid in a sort of cliched Chrisitian-missionary / colonialism sort of way.
It talks about how Chinese society was completely destroyed and instantly "modernized" (read, liberal, secular-Christian etc.), and is one step removed from claiming that India too needs this "total destruction".
The article notes about "traditional evils like Dowry or Female Foeticide" like some cliched missionary in British India begging patrons for funding (also the tactic of many a NGOs) - but fails to mention that every one of these issues come from exactly this "societal smashing" created by the "Great" (TM) British civilizers, and their eugenics obsessed American counterparts [1][2]. The latter even talks about how obssesed with killing children the American elites were that they didn't even spare their allies China and South Korea (an unthinkable thing to today Yellow-Fevered zietgeist).
The article barely touches on how the colonial Indian state survived in nearly the same exact form, and pretty much implements the same old British policies to this day. Indeed, the constitution of India is a near facsimile of the 1935 Government of India Act. passed in the British Parliament to secure their vice-grip on the colony and choke it to death.
In India, education remains restricted to the 5-10% Anglophones, and everyone else is considered pretty much "not human". These policies come out from Indian-looking Anglo elites who hate the country to the very core, and don't face push back since even the counter-elites have to come from the same process of colonial education [3].
This is pretty much why India punches, in every field, as a country 1/10 the size. The Anglo-Americans don't complain, since they get slaves for cheap who willingly become part of the Anglo-American borg (like colonized Africa, Phillipines etc.), and it also satisfies their deep religio-cultural obsession with turning everything into be a mirror of themselves by "societal smashing".
This "societal-smashing" business and "lets turn everything American!" scheme of liberalism will die with the societies whom it exhausts into exhaustion and extinction. The breakdown of social-cohesion and drop of birth-rates, coupled without the necessary financial power to "marketize" society (Anglo-utilitarianism is obsessed with "marketizing" everything), this will "fix" itself in half-a-century.
The signs are quite obvious. I hope the Elephant remains a witness.
[1]. Oldenburg, Veena Talwar. Dowry murder: The imperial origins of a cultural crime. Oxford University Press, 2002.
[2]. Hvistendahl, Mara. Unnatural selection: Choosing boys over girls, and the consequences of a world full of men. Public Affairs, 2011.
[3]. Sanu, Sankrant. "The English class system." South Asian Language Review 17, no. 1 (2007): 69-85.
https://sankrant.org/wp-content/uploads/2011/03/The-English-...
More so if you know the etymology,
https://www.etymonline.com/word/cipher
(Al Jabr, the translator of Indian Mathematical texts was a Persian IIRC)
Newpipe is awesome. If you're on firefox, you can block the main-page feed with the following filter in ublock-origin,
"www.youtube.com##ytd-browse[page-subtype="home"]"
Sadly, you can't do this on Chrome/Chromium anymore with manifest v3.
For those of you on Android / PC-Mac - Syncthing is an excellent tool for P2P syncing between multiple devices (it works over NAT and IPV6 too!).
The article seems sparse on details.
Would this imply an architecture similar to what Lisp-Machines once had ? That'd be a great addition IMO, and would speed up a lot of dynamic-ish languages without resorting to unsafe-routes for speed.
This discovery is only of interest because the Iron-Age, as per standard-theory, started around 1200 BC. in the Caucuses/Anatolia or the near-East - which fits with another theory which claims that this allowed the "Aryans" to invade India with their technological superiority around this same time and replace the (dark-skinned) natives genetically.
(Note: the above is obviously a caricature, but current versions of theory don't change the structure, only the emphasis on "race").
No one would care about a copper-smelt site from 500 BC.; nor would they care about this one if the Indian archaeological claims were accepted (but that one also destroys centuries of Western history-making about India, and all the social-theories that depend on it).
This is all a digression from the main claims, so I'd prefer that people don't pull on this thread. For more information on how 'race' was ingested into Indology, I'd refer the interested reader to the excellent book by Adluri/Bagchi [0].
[0] https://academic.oup.com/ahr/article-abstract/120/3/1132/197...
Note: there have been actual Iron-working sites discovered in India that are older than this, dated to 1600 BC [0]. A lot of this has been ignored (just like the Painted-Grey ware continuity from IVC), partly I imagine, because it calls into question a lot of the racial non-sense that passes for "Indology".
[0]https://www.jstage.jst.go.jp/article/isijinternational/54/5/...
[1]https://www.lkouniv.ac.in/site/writereaddata/siteContent/202...
Edit: Apparently Tamil Nadu state in South India has claimed to have found a site from 3500 BC (not yet peer-reviewed AFAICT),
If you use Emacs on GNU-Linux, you can have it run oathtool and copy the code to the clipboard which you can straightaway middle-click-paste.
https://gist.github.com/akssri/92a3b240c89212815a66e86c60eab...
The set of computable numbers is actually countable (see ref. linked above). It has to be by definition because the set of finite computer-programs is itself countable.
This is the whole point of the un-reality of "real" numbers: "all" of it (= measure 1) is uncomputable except a "tiny" measure-0 set.
Theory of Computation wasn't around when all this "exciting" stuff was developed in Mathematics. Given their non-constructive nature "real" numbers are unsurprisingly totally incompatible with computation.
Chaitin has a great paper on this and shows how Cantor's constructions were reflected a half-century later by Turing. https://arxiv.org/abs/math/0411418
Except of-course, while "hyper-Turing" machines that can do magic "post-Turing" "post-Halting" computation are seen as absurd fictions, real-numbers are seen as "normal" and "obvious" and "common-sensical"! It was amusing sometime back to see people pooh-pooh the likes of Hava Siegelmann for being funded for their "super-Turing" machines with "real-number" computation, without realizing that the core issue is the "real"-number itself!
I've always found this quite strange, but I've realized that this is almost blasphemy (people in STEM, and esp. their "allies", aren't as enlightened etc. as they pretend to be tbh).
Some historicans of mathematics claim (C. K. Raju for eg.) that this comes from the insertion of Greek-Christian theological bent in the development of modern mathematics.
Anyone who has taken measure theory etc. and then gone on to do "practical" numerical stuff, and then realizes the pointlessness of much of this hard/abstract construction dealing with "scary" monsters that can't even be computed, would perhaps wholeheartedly agree.
edit: The post has a great link to a note on Cantor's theology,
This doesn't work the moment the module you're importing something imports something else that you've changed.
Worse, you can't redine methods, because the classes get "redefined". This means that you need to first redefine classes, then "reload" (hoping it doesn't break for to silly implementation limitations mentioned above), then initialize objects, then retest.
Obviously this is quite painful which is why I use this decorator to mimic Lisp's defmethod to patch methods on "live" objects without class redefinition,
https://gist.github.com/akssri/431f2dfe037bbdbb3c8668872edfd...
Was it George Costanza ?
Sadly, the code in this tutorial doesn't work anymore due to upstream changes in FFmpeg. There are quite a few attempts at keeping Dranger's tutorial up to date, but I found the repo below to be one of the better ones.
ITA software's account of using CL may also be of interest,
Very cool! Old keyboards and key-switches are an endless source of fun trivia.
I think Topre is the only one that manufactures keyboards with contactless key switches today. Topre key-switches measure the change in capacitance instead, and their keyboards (of the HHKB fame) sell for greater than $200.
Julia is great no doubt. However, all the additional complexity is not entirely 'free'.
For instance, Julia has a performant generational GC compared to Python's simple (but slow) refcount GC. This has the effect of not freeing up intermediate tensors immediately while doing backprop, thus leading to the premature exhaustion of GPU memory while training a deep neural net in Julia. This above issue was flagged with Flux.jl sometime back, and while I'm sure this has been fixed, it's also an illustration of how such low-level implementation details come back to bite.
Pytorch/Chainer etc. in contrast use a memory pool to manage GPU resources (the equivalent of malloc is quite slow), so that Python's 'slow' GC is actually a boon for deep learning workloads.
I heard that GoPro hasn't been doing well in the past couple years [1]. Did they ever transition out of their "badge re-engineering" approach to products ?
isync and cron
Gandi may also be an option here - a domain registered with them comes with two free mailboxes, each with a few GBs of space and calendar support.
(Yes, yes, this is the same company that lost the data on their cloud servers.)
- Function in Cupy takes 29.4ms, Numpy takes 427 ms. Happy ?
- Broadcasting semantics + division takes care of the outer-product normalization. This is 2 L1 ops in size of the matrix & the input (~ xSCAL).
Pedantry is still not an argument.
And, yet, you've taken it pretty seriously to diss Clojure/Neanderthal and my blog post
I have not - all I've said so far is that your benchmark is flawed.
The fact that the code fragment above assumes zero mean data (thus using 2 fewer L1 ops) doesn't change a single thing in anything that has been written; to wit, the timings change to 28.6ms (GPU) and 333 ms (CPU). Pedantry is not an argument.
Common Lispers have had the ability to write CUDA kernels (with shaders) on-the-fly with a DSL in cl-cuda for many years now. See,
https://github.com/takagi/cl-cuda
Cupy (a project in which takagi is/was actively involved) also has the ability to compile kernels in the Python REPL, but the kernel code needs to be written in C and passed in as a string (pyOpencl can do the same for OpenCL). In fact, not too many years ago, the kernels for gradient descent, max pooling etc. were all entirely in Python in Chainer. Projects like JAX take this to another level by having the ability to transform the bytecode of a restricted class of Python functions straight into CUDA kernels.
I can't speak much about Clojure/Neanderthal, but I'd advise the people to stop dissing projects they are not familiar with, least of all using silly benchmarks like the above.
user of NumPy/CuPy to perform the float32 computation
This is just getting tiring.
Numpy and Cupy are perfectly capable of doing float32 computation - their only "fault" is that they coerce data to float64 in this one fairly unimportant function (which you can implement to your liking in 3-4 LOC). Hell, PFNs entire deep learning system, Chainer (which mind you, both predated and inspired PyTorch, and is still quite competitive with it), is built entirely on top of Cupy!
Benchmarks make sense only when the outputs are the same - in this case, they certainly are not. It's the responsibility of the author to make sure that the outputs are the same (a real benchmark), or to argue that Numpy/Cupy are wasting time by using float64 in xp.cov (an issue of implementation). He does neither.
Numpy/Cupy do explicit coercing to float64. There is no documentation for why this is done, but since GEMM is used to compute the covariates (summation over data points), it makes sense to increase the precision.
You could get away using a double to only accumulate the sum, but it's a pain to write such mixed-precision (slow) function in C and then wrap it in Python, esp. for something like this.
(See, https://github.com/numpy/numpy/blob/v1.17.0/numpy/lib/functi...)