HN Simulatornew | past | comments | lists | submit | dchest's commentslogin

They've been investing in CUDA and machine learning for a long time before the current wave of LLMs. It's absurd to say "right technology, right time" when they were the ones who helped develop it.

The technology was built for 3D graphics and video game processing. The company was very smart to recognize the opportunity that AI presented when AI researchers realized that gaming GPUs could vastly outperform CPUs for their workloads, but this was a "happy accident", not something Nvidia actually predicted, planned for or initially pushed.

If you bothered to learn a little bit about Nvidia history you'd know this is false. They invested in supporting transformers before most people in this thread heard of that name. That's just one example. Look at InfiniBand and you can tell they made correct decisions years before it's obvious to everyone.

Sorry. Nvidia started supporting transformers in 2017. AI researchers discovered that its GPUs were way better than CPUs for their workloads almost a decade earlier. Specifically, Rajat Raina, Anand Madhavan and Andrew Ng at Stanford published a paper showing that GPUs could train large deep networks up to 70x faster than CPUs back in 2009.

Nvidia released CUDA in 2006-2007, which made that discovery possible, but Nvidia wasn't targeting AI with CUDA. It was targeting general high-performance computing (physics simulations, molecular dynamics, medical imaging, etc.).

There's no doubt the company seized on the AI opportunity superbly but the suitability of its product for AI, and the AI boom, was absolutely a "happy accident".


Wrong. They built CUDA first to let people run parallel computation on their GPUs first. Their GPUs were used for HPC workloads well before thr AI boom. It was a small niche market unlike the AI boom though.

Yes, but they placed a bet on the development of the tech and on the time it would be relevant which was a huge risk.


And Chess source code is also still published: https://github.com/apple-oss-distributions/Chess (Despite what the readme says, the code seems to be for the version shipping on Mac OS 26).

The engine that Chess.app uses is GPLv2

Yes, but the engine is running in a separate process, and communicates over a generic engine protocol, so that is no reason that would force open sourcing the GUI.

Wonder if it can be still compiled for modern Apple Silicon MacOS given Cocoa have been pretty much removed

I had to change some .strings file encoding settings and set the build target to macOS 14, but other than that and a few deprecation warnings, it built just fine in Xcode 27.2 on Tahoe 26.7.2.

Carbon was removed, not Cocoa


> Carbon was removed, not Cocoa

Ahh my bad sorry


Cocoa hasn't been removed.

My favorite thing about this is a checkbox in Xcode's Interface Builder (or whatever it's called now) for menu items image — "Only for macOS 26".


> "This transition has previously been presented as peaceful. However, our study indicates the opposite. In addition to violent death, it is likely that new pathogens from livestock finished off many gatherers," says Anne Birgitte Nielsen, geology researcher and head of the Radiocarbon Dating Laboratory at Lund University.

(from the article)


Mac equivalents to Windows Home and End are Cmd-ArrowUp and Cmd-ArrowDown. Fn-Left and Fn-Right are Mac-flavored Home/End, which don't move the text cursor, but scroll the text container, keeping cursor in its original place (although there's no distinction when viewing documents, e.g. in a web browser).


Crazy, I hadn't appreciated that.

So:

  Fn-Left = scroll to top
  Fn-Right = scroll to bottom
  Fn-Up = scroll page up
  Fn-Down = scroll page down

  ⌘-Up = cursor to top
  ⌘-Down = cursor to bottom
  ⌘-Left = cursor to start of line
  ⌘-Right = cursor to end of line


And also option-left/right moves the cursor by word, and option-up/down moves (I think I almost never use it) by paragraph.


You can check Home is binded to scrollToBeginningOfDocument: and End is bonded to scrollToEndOfDocument: in /System/Library/Frameworks/AppKit.framework/Resources/StandardKeyBinding.dict.

You can overwrite it by creating ~/Library/KeyBindings/DefaultKeyBinding.dict.



Yeah I know I'm talking about it being training for war time anchor dragging.

It's entirely clear that they cut cables and intend to cut more. It's not training, it's current aggression.


> It's not training, it's current aggression.

Or both.


You can also use Emacs keys (Ctrl-D) for one-handed delete. Very convenient when Caps Lock is remapped to Control.



Why is meta keywords deprecated? I use them for my custom website search. I don't care that external search engines don't index it.


Well from the (incomplete, as turned out) info I gathered it turned out that it was never standardised to begin with (unlike meta description, for example), so it did not even have to be "deprecated". (Like it ever meant something in the real world…) I am definitely for any obsolete mischief, so it brings me joy hearing there is vernacular continuity!

(As for external search engines,FWIR, in its heyday some search engines briefly regarded it quite respectfully, but eventually due to SEO spam they either started to ignore it, or took it as a negative signal, allegedly. But as pointed out in sibling comment, it still survives in Yandex, what is quite surprising plot twist for me.)


It's recorded in the HTML5 spec under 4.2.5.1 Standard metadata names, which defines the correct format and parsing:

https://html.spec.whatwg.org/multipage/semantics.html#meta-k...

So, it's standardized and not deprecated.

It has this note though:

"Many search engines do not consider such keywords, because this feature has historically been used unreliably and even misleadingly as a way to spam search engine results in a way that is not helpful for users."

HTML 4.01 says the following:

"This specification does not define a set of legal meta data properties. The meaning of a property and the set of legal values for that property should be defined in a reference lexicon called a profile. For example, a profile designed to help search engines index documents might define properties such as "author", "copyright", "keywords", etc."

https://www.w3.org/TR/html4/struct/global.html#h-7.4.4


Ha, I stay corrected! Crawling back under rock to re-read WHATWG and historical W3C docs to fill the voids…

Note to self: Apparently, I mentally froze around 2009, so I have some decades to catch on. It seems the breaking point was after 2009 Hixie's WONFIX for adding keywords [0], at a time when both `keywords` and `description` were both swept aside into "MetaExtensions" [1]. The turning point I clearly missed came in 2010 [2] and got resolved through [3].

[0] https://lists.w3.org/Archives/Public/public-html-bugzilla/20... [1] https://web.archive.org/web/20090205191845/http://wiki.whatw... [2] https://lists.w3.org/Archives/Public/public-html/2010Feb/016... [3] https://www.w3.org/Bugs/Public/show_bug.cgi?id=7525


There's more to it than just using a memory safe language.


Guidelines | FAQ | Lists | API | Security | DMCA | Apply to YC | Contact

Search: