Hacker News: rs38

New comment by rs38 in "Qwen 3.8 27B"

rs38 — Fri, 14 Aug 2026 19:17:42 +0000

pretty sure this exists already...

New comment by rs38 in "Gemini Distillation Service"

rs38 — Wed, 29 Jul 2026 06:31:11 +0000

404?

New comment by rs38 in "Kimi-K3 on HuggingFace"

rs38 — Mon, 27 Jul 2026 08:07:01 +0000

is there a realistic way to distill 2 consumer hardware friendly models with max ~200B and ~20B? Qwen did it, but would it be possible for 3rd parties (unsloth etc)?

New comment by rs38 in "Local Qwen isn't a worse Opus, it's a different tool"

rs38 — Thu, 18 Jun 2026 18:34:27 +0000

wanna chime in, recently tried vLLM to consume a NVFP4 Gemma4 safetensor model and see how the batching can show up in nice t/s numbers. it's slow to start, it's Linux only, it doesn't like WSL much, ended up with either old or nightly container builds, I more or less have given up. Appreciate how llama.cpp simply works and does things fast and obvious

New comment by rs38 in "Running local models on an M4 with 24GB memory"

rs38 — Mon, 11 May 2026 12:49:09 +0000

my latest experiments with local LLM (mistral coder variations) fitting in older 6 GB GTX1060 were disappointing as long as you try to hook Copilot (CLI or VScode) to it and are used to provide a lot tooling. this seems to bloat initial prompt to 20k and more which seems the bottleneck if I did not completely misconfigured things. output tokens/s are more than fine, but PP is frustrating / unusable.

New comment by rs38 in "GitHub Copilot is moving to usage-based billing"

rs38 — Mon, 27 Apr 2026 20:53:26 +0000

both 5.4 were best bang (-mini even more as I found it usually same well performing!!!) for the buck before. Now they face cost reality it seems.

New comment by rs38 in "GitHub Copilot is moving to usage-based billing"

rs38 — Mon, 27 Apr 2026 20:49:25 +0000

e.g. if on an annual plan? 0x will be gone, but there are okay 1x and 0.3x models left. I am pretty much curious how the early may test invoicing will look like. current setup of tools etc. is way too chatty eats up 1+M token per PRU easily. not sure how much is cached.

New comment by rs38 in "Meta replaces WhatsApp for Windows with web wrapper"

rs38 — Thu, 13 Nov 2025 15:38:14 +0000

not sure how much resources it took from my WinARM Win11 as store app, but it's buggy in either release or beta since months as it loses network connectivity after some time...

New comment by rs38 in "[dead]"

rs38 — Fri, 07 Nov 2025 07:30:56 +0000

I think it's more Dave that "sucks"...click baiting titles, no need to and ever and ever repeating that he was at MS. He should do Podcasts as there is no value in videos only speaking (feels like 1.5x btw)

New comment by rs38 in "Ask HN: In what ways has AI tools have replaced your old way of doing things?"

rs38 — Tue, 04 Mar 2025 07:24:25 +0000

Stackoverflow usage down to 10%

New comment by rs38 in "Nvidia's Project Digits is a 'personal AI supercomputer'"

rs38 — Wed, 08 Jan 2025 07:23:51 +0000

at least the naming implies factor ~20 less :)

New comment by rs38 in "Frontier Models are Capable of In-context Scheming [pdf]"

rs38 — Tue, 10 Dec 2024 11:55:10 +0000

this paper is going through mass media meanwhile as it alledgely shows how Skynet started to take over...

New comment by rs38 in "55 GiB/s FizzBuzz (2021)"

rs38 — Mon, 03 Jul 2023 06:59:49 +0000

found my glasses and there is a candidate AVX optimized to 1 GB/s

New comment by rs38 in "55 GiB/s FizzBuzz (2021)"

rs38 — Mon, 03 Jul 2023 06:48:59 +0000

do I need glasses, or not a single c#/.net implementation?

New comment by rs38 in "Ask HN: How to learn sales?"

rs38 — Mon, 28 Sep 2020 06:43:39 +0000

nice one:

https://www.youtube.com/watch?v=ejA0wKTRHH4

Presentation shown to new Mercedes salsman in the late 1980s and early 90s. Covers lots of sales tips and techniques to help put the customer in the driver's seat of a new Mercedes.

New comment by rs38 in "Show HN: Avoidting – Avoid meetings by appearing busy"

rs38 — Wed, 04 Jan 2017 14:50:35 +0000

great!