Viewing profile — slashdave
slashdave
HN member- Joined
- Wed, May 08, 2024, 9:42 PM UTC
- HN karma
- 767
- Public activity
- 657 items
- HN profile
- View on Hacker News ↗
About slashdave
No profile information was provided.
Recent public activity
-
comment
Comment #49225689
M4's GPU does not support 64 float. So, yeah. No hardware support, so what are you expecting? The fact that it comes close is pretty impressive.
-
comment
Comment #49172572
I mean... who has control over OpenAI's web page? No self-respecting PR firm would do something like this. And Counsel must be hopping mad.
-
comment
Comment #49172556
Please don't assume the entire AI industry behaves like OpenAI
-
comment
Comment #49172538
Why do think they would want to be vindictive? The point is to prevent leaks, leaving lapses is exactly counter to that goal.
-
comment
Comment #49172503
It's not hard if you are honest
-
comment
Comment #49164785
Those that claim their opponents are in denial are most likely in denial themselves
-
comment
Comment #49164765
Since when does realism become hope? (It's the other way around)
-
comment
Comment #49164760
Shh! Don't upset the hype train
-
comment
Comment #49164575
No, we do this kind of thing regularly. Optimizations often depend on the model that is using them. Namely, fusion techniques.
-
comment
Comment #49151343
Indeed. Compilers do have to make compromises, and it is not possible from code alone to understand the tradeoffs.
-
comment
Comment #49151333
This SV mentality that technology can solve all problems is rather tiring, really.
-
comment
Comment #49139974
Have your agent take care of it
-
comment
Comment #49137892
> We only know how to do it by means of considerable sacrifice Little sacrifice actually > Solving the issue would be doing it without sacrifice So... you are expecting magic? LLMs…
-
comment
Comment #49136700
Yeah, a corner case (streaming) and good developers working very hard. The original SSE (SIMD) extensions were actually written with this use case in mind. When they first came out…
-
comment
Comment #49136026
Except for some odd corner cases, it is very hard for a good engineer to beat compilers these days (they are very good now). Not to mention, tuning depends on the platform, and oft…
-
comment
Comment #49135994
Show it the disassembler...
-
comment
Comment #49135978
I'm... confused. Don't we use macro assemblers anymore?
-
comment
Comment #49135959
Is AWS listening?
-
comment
Comment #49135925
Seriously? We already know how to solve all of these issues. What we lack is collective political will.
-
comment
Comment #49135917
What? No. Frontier physics is experiment driven.
-
comment
Comment #49135905
It will not happen with existing LLM techniques.
-
comment
Comment #49135872
> we are also seeing incredible advances in software performance Incredible? > open ai announced like 15% improvement by fixing gpu kernel issue That is... ordinary software optimi…
-
comment
Comment #49135838
> The sooner people can be broken out of their denial There is irony here
-
comment
Comment #49135791
Cursor achieved fast adoption by making it simple to move from Visual Studio code. Double-edged sword. It is also simple to move back to VS code and agent extensions.
-
comment
Comment #49117892
Don't forget: resources