sanxiyn

19 ms·

241.
by sanxiyn 3y ago
After years of trying (IIRC there was funded effort from Intel), LLVM still can't parallelize code generation at function level. Cranelift had function level code generation parallelization from day 1.
242.
by sanxiyn 3y ago
Windows is supported. See https://github.com/rust-lang/rustc_codegen_cranelift/issues/... .
243.
by sanxiyn 3y ago
It is done in the shared code, see https://rustc-dev-guide.rust-lang.org/backend/backend-agnost... for details.
244.
by sanxiyn 3y ago
In theory yes, this could be problematic with alternative Rust implementations, but it works fine and many people are using it in this case. We can be sure LLVM and Cranelift codegen backends will be implementing the same binary ABI, becaus
245.
by sanxiyn 3y ago
Have a look at Unison which uses constraint programming to do optimal code generation. https://unison-code.github.io/
246.
by sanxiyn 3y ago
I too was impressed at first, but got disillusioned after learning how they did it. It was much more similar to chess AI with piece-square tables than I first thought. https://gist.github.com/dfarhi/66ec9d760ae0c49a5c49
247.
by sanxiyn 3y ago
This is as bad as China banning YouTube. US fell to the level of China.
248.
by sanxiyn 3y ago
Error analysis of cosmic distance ladder is fiercely technical subject. https://arxiv.org/abs/1103.2976 Table 5 is titled "H0 Error Budget for Cepheid and SN Ia Distance Ladders". (It is old and the field is
249.
by sanxiyn 3y ago
This in fact happened to Cloudflare with Cloudbleed and their decision was to switch to Rust. There is a human factor at work here. Rationally speaking, Herb is right 98% reduction is sufficient. But when the CTO's head is on the line,
250.
by sanxiyn 3y ago
S-expression still needs a schema. Just as you can't say "use XML", neither you can say "use S-expression". Most of work of defining DocBook is about defining DocBook schema, not about defining DocBook XML syntax.
251.
by sanxiyn 3y ago
Since this was well received, here is a full translation of The Song of Hyundai Heavy Industries' Labor Union, done by ChatGPT and reviewed by me: The Song of Hyundai Heavy Industries Union Brothers of Hyundai, standing tall,
252.
by sanxiyn 3y ago
I am from South Korea. Hyundai Heavy Industries' Labor Union, the union of the largest and probably the most productive shipyard in the world in Ulsan, is likely one of the strongest union in the world. The union maintains a YouTube ch
253.
by sanxiyn 3y ago
If you think https://arxiv.org/abs/1801.01203 is a good paper, I am not sure why this is any different. Yes, we want a paper for every way the technology can be subverted.
254.
by sanxiyn 3y ago
It is concerning, but I am not sure whether it is more concerning than that it's so hard to write a web browser that doesn't execute arbitrary code. Security is like that, and security is especially hard when the system is feature
255.
by sanxiyn 3y ago
No, donors have no standing. See https://www.thetaxadviser.com/issues/2021/sep/donor-no-stand... .
256.
by sanxiyn 3y ago
I am unsure. You can't (for example) fine tune over API. Is anything safe for Microsoft to fine tune really safe for Russia, CCP, etc. to fine tune? Open weight (which I think is more accurate term than open source here) models enable
257.
Why are all proteins left-handed? New theory could solve origin of life mystery(science.org)
4 points by sanxiyn 3y ago|0 comments
258.
by sanxiyn 3y ago
Isn't Norway operating under a form of capitalism as well as owning controlling share of Statoil? Why mention China or Russia? I guess Norway is not a member of EU.
259.
by sanxiyn 3y ago
I thought EU had money a lot more than €15M (Microsoft investment), actually a lot more than €1.8B (Mistral valuation). Why didn't EU make an offer? If EU wants to own its AI future, it has enough money to buy one. They decided not to.
260.
by sanxiyn 3y ago
https://lobste.rs/s/0akqy7#c_ihruks
261.
by sanxiyn 3y ago
Decision to abandon C/C++ is okay I would say. Decision to abandon C/C++ immediately is less so.
262.
by sanxiyn 3y ago
These kinds of top down directives are almost always wasteful, because they ignore local situations. It could have been case by case decisions, considering cost and benefit. Now I understand why this is politically difficult, but it still r
263.
by sanxiyn 3y ago
I heard that when Cloudbleed happened, Cloudflare CTO declared that there will be no more new projects in C or C++, and that's how it happened. Needless to say, lots of time and money were wasted.
264.
by sanxiyn 3y ago
I am somewhat skeptical about high-NA EUV. There are reasons why Intel got priority on that: ASML marketing materials themselves imply high-NA EUV is worse than low-NA EUV in terms of cost. https://www.semianalysis.com/p&#x2
265.
by sanxiyn 3y ago
"Berkeley is already pretty dense" is a laughable statement with its single-family homes near downtown.
266.
by sanxiyn 3y ago
You can also batch requests using different LoRAs. See "S-LoRA: Serving Thousands of Concurrent LoRA Adapters". https://arxiv.org/abs/2311.03285
267.
by sanxiyn 3y ago
You can use outlines https://github.com/outlines-dev/outlines to let models generate with correct syntax.
268.
by sanxiyn 3y ago
> 25% code coverage improvement The paper does not report code coverage improvement and it is probably not 25%. The paper does say this: > The median number of lines of code added by a TestGen-LLM test in the test-a-thon was 2.5.
269.
by sanxiyn 3y ago
25% of test cases increased coverage, not coverage was increased by 25%. For example, if they started with 90% coverage and each test case increased coverage by 0.1% and there were 100 test cases, final coverage was 92.5%.
270.
by sanxiyn 3y ago
Well, yes, but in the US USPTO is responsible for both trademarks and patents.

More ›