This is an understatement: the “source” for an LLM or similar deep learned network is not the weights, but the training data and the scripts required to train it from scratch.
There’s a great paper from the group of Emery Berger, where they merge memory pages that become sparse. A kind of garbage collection for C and C++ languages. There’s also a good presentation about it on youtube.
It’s a billion times integer modulus, summed. Div/mod is by far the slowest arithmetic instruction on a CPU, so it underestimates C/rust performance. I guess the author chose that to prevent the C compiler from optimizing the loop away.
These extremes exist, and GR predictions are better than Newton’s in those cases. Closest to home is mercury’s perihelion drift. We have observed black hole mergers, gravitational lensing, and GR is also an essential component in understanding the universe’s expansion(that we know from redshift and the CMB). Likely MOND will address these, but Newtonian mechanics will not get you there.
The way I thought LCD/LED displays worked was by RGB filtering a uniform white backlight. Is it only this design that does fosforescence per subpixel? Sounds way more energy efficient.
It depends on the application. It’s the easiest way especially if you might encounter right to left script, CJK or emoji. It is worthwhile to cache the textures, most text does not change every frame. It is good enough for us.
I would argue it is conceivable that given enough time, a monkey or ape species can evolve that is intelligent enough to write Shakespeare’s full works.
The RAM is expandable as well… however I am curious how well the extra RAM performs. Part of the M-series performance gain is from having the RAM dies very close to the processor.
An example for those not in the know: to find a shortest route on a realworld map, an admissible heuristic would be that the minimum travel distance between two nodes will be a straight line. While examining options, A* takes this into account, Dijkstra does not.
As an alternative to simulating latency: How about using a VPN service to test your website via Australia? I suppose that when it easier to do, it is more likely that people will actually do this test.
The author wonders why he does not see a memory footprint for the billions of objects, and conjectures they are stored in-line in the array. I suspect that it is actually achieved by the flyweight pattern, where copies of the same immutable (reference type) object are simply the same object.
I have tried the windows version 2 years ago. The site I copied was our on-prem issue tracker (fogbugz) that we replaced.
HTTrack did not work because of too much javascript rendering, and I could not figure out how to make it login.
What I ended up doing was embedding a browser (WebView2) in a C# Desktop app. You can intercept all the images/css, and after the Javascript rendering was complete, write out the DOM content to a html file.
Also nice is that you can login by hand if needed, and you can generate all urls from code.
It annoys me too, especially when the explanation is: “because RISC simpler so it is faster, duh”. However x64 is at disadvantage for instruction decoding (it’s difficult to decode multiple variable length instructions in parallel), and secondly the guarantees for memory coherence across cores. Both come with an extra burden intel can’t eliminate for backward compatibility.
Such an experiment was described in a science book I read as a kid. They dissolved oil in alcohol so as to measure a very small quantity of oil, and then a a knitting needle was used to stretch the oil film across a plastic container filled to the edge with water (until the film breaks).
How about applying the idea behind ESCROW: if you market hardware with software dependencies, you are required to provide the source to a trusted third party who will release/opensource it if you stop maintaining said software before the expected lifetime of the hardware.
On a motorway, 50 to 80 eurocents per kWh is what I often see. There is not enough competition.
It is still cheaper than what you’d spend on fossil fuels.
However, for day to day use you charge at home or at the office which is much cheaper (maybe 25 cents?).