Hacker News: flamedoge

New comment by flamedoge in "The path to ubiquitous AI (17k tokens/sec)"

flamedoge — Sat, 21 Feb 2026 07:02:27 +0000

I have a gut feeling, huge portion of deficiencies we note with AI is just reflection of the training data. For instance, wiki/reddit/etc internet is just a soup of human description of the world model, not the actual world model itself. There are gaps or holes in the knowledge because codified summary of world is what is remarkable to us humans, not a 100% faithful, comprehensive description of the world. What is obvious to us humans with lived real world experience often does not make it into the training data. A simple, demonstrable example is whether one should walk or drive to car wash.

New comment by flamedoge in "Furiosa: 3.5x efficiency over H100s"

flamedoge — Thu, 15 Jan 2026 10:30:04 +0000

it's so weird how they spend all this money to train new models and then open sources it. it's gold rush but nvidia is getting all the gold.

New comment by flamedoge in "LLVM: The bad parts"

flamedoge — Tue, 13 Jan 2026 11:51:32 +0000

non-C/C++ centric, performant compiler maybe. Aliasing support in C is pretty limited and a performant langauge like fortran and more modern equivalents may seek more efficient, concise IR for compiler with less comparable overhead from LLVM.

New comment by flamedoge in "The C++ standard for the F-35 Fighter Jet [video]"

flamedoge — Mon, 08 Dec 2025 06:26:45 +0000

surprised not that strict about types. i remember google doesn't allow unsigned?

New comment by flamedoge in "Fp8 runs ~100 tflops faster when the kernel name has "cutlass" in it"

flamedoge — Fri, 03 Oct 2025 15:31:35 +0000

Let's be clear here. Intel searched for "GenuineIntel" and ran optimized code on their hardware.

New comment by flamedoge in "Writing an operating system kernel from scratch"

flamedoge — Sun, 14 Sep 2025 18:41:07 +0000

To you maybe. The subset of population that is even interested, smart, persevering to do this is extremely tiny.

New comment by flamedoge in "U.S. government takes 10% stake in Intel"

flamedoge — Fri, 22 Aug 2025 23:54:09 +0000

so.. shouldn't US take stake in TSMC instead?

New comment by flamedoge in "Ask HN: Any insider takes on Yann LeCun's push against current architectures?"

flamedoge — Sat, 15 Mar 2025 09:06:56 +0000

It literally doesn't know how to handle 'I don't know' and needs to be taught. fascinating.

New comment by flamedoge in "OpenAI O3-Mini"

flamedoge — Sat, 01 Feb 2025 18:45:16 +0000

Soon people will read other people's summary that they copied from AI summary on the web.

New comment by flamedoge in "Amazon workers to strike at multiple US warehouses during busy holiday season"

flamedoge — Thu, 19 Dec 2024 09:26:49 +0000

Order from local store, order from Amazon, use first one to receive, return new one to whichever was more expensive.

New comment by flamedoge in "Meta Movie Gen"

flamedoge — Sat, 05 Oct 2024 07:02:16 +0000

who's stopping you from Amish lifestyle? problem seems to be that people want authentic, hand-made 'art' but with price of a mass manufactured tech.

New comment by flamedoge in "NASA acknowledges it cannot quantify risk of Starliner propulsion issues"

flamedoge — Mon, 19 Aug 2024 18:25:13 +0000

so whats it good for. sending monkeys?

New comment by flamedoge in "Noexcept affects libstdc++’s unordered_set"

flamedoge — Sun, 18 Aug 2024 18:38:08 +0000

what.. noexcept throws exception..? what kind of infinite wisdom led to this

New comment by flamedoge in "CrowdStrike Official RCA is now out [pdf]"

flamedoge — Wed, 07 Aug 2024 22:25:52 +0000

regex isn't probably a good thing in a kernel boot code considering it's NP hard

New comment by flamedoge in "C++'s `noexcept` can sometimes help or hurt performance"

flamedoge — Tue, 06 Aug 2024 08:13:22 +0000

> did not restrict optimization unnecessarily.

well clearly there is a cost

New comment by flamedoge in "Zen 5's 2-ahead branch predictor: how a 30 year old idea allows for new tricks"

flamedoge — Sat, 27 Jul 2024 05:53:03 +0000

I wonder what they need before this change. Branch predictor hardware may not have accounted for depth beyond single conditional branch? but pipeline was probably always filled, unpredicted.

New comment by flamedoge in "Run CUDA, unmodified, on AMD GPUs"

flamedoge — Tue, 16 Jul 2024 23:33:57 +0000

probably because their focus is on data centers that mostly run linux

New comment by flamedoge in "Beating the Compiler"

flamedoge — Sat, 13 Jul 2024 13:23:43 +0000

and well worth the cost

New comment by flamedoge in "Intel pauses work on $25B Israel fab"

flamedoge — Tue, 11 Jun 2024 03:54:57 +0000

not so baffling when you realize those places aren't blessed with natural resources. people have to use their brainpower for gdp there

New comment by flamedoge in "The Google employees who created transformers"

flamedoge — Thu, 21 Mar 2024 03:52:42 +0000

Maybe. I can't be the only one who gets irked by the slow, typing-out response from copilot/gpt when google gets me instant results. Near instant or a lot faster result might convince me to use whatever gpt search