Hacker News: potus_kushner

New comment by potus_kushner in "Show HN: I shrank DeepSeek V4 Flash to 57GB and it wrote a compiler on my Mac"

potus_kushner — Sun, 16 Aug 2026 22:16:59 +0000

impressive work. but a GGUF release + (ik_)llama.cpp PR would be appreciated since not everybody has a mac.

New comment by potus_kushner in "Dots3-Note Preview"

potus_kushner — Sat, 15 Aug 2026 14:44:19 +0000

pretty impressive model for its size, if the benchmarks can be trusted.

New comment by potus_kushner in "Qwen 3.8 27B"

potus_kushner — Fri, 14 Aug 2026 22:45:46 +0000

hopefully for us mere mortals without $4k+ hardware a 35B MOE model will be released. or a new prism ternary bonsai model based on this one.

New comment by potus_kushner in "To save C, we must save ABI (2022)"

potus_kushner — Wed, 12 Aug 2026 00:47:19 +0000

the ramblings of a madman.

New comment by potus_kushner in "Linux manpages maintainer wants to raise memory.h from the dead"

potus_kushner — Tue, 04 Aug 2026 01:17:49 +0000

memcpy and friends have been living peacefully in string.h since 1989, but alejandro colomar wants to move them to memory.h.

since he's overtaken the maintainance of the linux manpages project, his unilateral moves to change libc interfaces according to his personal preferences have been confusing and infuriating. his work is to document existing interfaces, but instead he abuses his position to override existing specifications, in what appears to be acts of sabotage.

Linux manpages maintainer wants to raise memory.h from the dead

potus_kushner — Tue, 04 Aug 2026 01:17:49 +0000

Article URL: https://lore.kernel.org/linux-man/am0RdPxvbIYUKAL-@devuan/T/#m277b916c39713d644d36a42d82debbff1df6bfc5

Comments URL: https://news.ycombinator.com/item?id=49163346

Points: 3

# Comments: 1

New comment by potus_kushner in "Demand from AI data centers drives up computer memory prices"

potus_kushner — Mon, 03 Aug 2026 21:59:47 +0000

AI is the perfect scapegoat. nobody talks about the price fixing. https://web.archive.org/web/20260701064235/https://www.polyg...

New comment by potus_kushner in "RipGrep musl binaries occasionally segfault during very-large searches"

potus_kushner — Sun, 02 Aug 2026 00:56:53 +0000

this use of musl's mallocng actually lead to the discovery of a kernel bug, thanks to its hardening. without it, this might have gone unnoticed for months, silently corrupting memory in the meantime.

AntGroup releases Ling 3.0 Flash MoE model

potus_kushner — Wed, 29 Jul 2026 18:30:42 +0000

Article URL: https://twitter.com/AntLingAGI/status/2080351022028095681

Comments URL: https://news.ycombinator.com/item?id=49101228

Points: 4

# Comments: 0

New comment by potus_kushner in "Prompt Caching in Agents"

potus_kushner — Fri, 24 Jul 2026 02:31:24 +0000

good article, killed by fancy animations driving all my cores to 100% cpu and fans to full noise-level, like a cryptominer. closed for that reason.

New comment by potus_kushner in "You only need the frontier model for one single edit"

potus_kushner — Tue, 21 Jul 2026 05:29:12 +0000

this article is full of missing text like this: " of Opus at of the cost, the speed, points over oneshot Flash." is it just me or is that page using some brand new js feature or chromium-only hack?

New comment by potus_kushner in "Solod: Go can be a better C"

potus_kushner — Sat, 18 Jul 2026 21:57:12 +0000

it's already happening: https://github.com/rolandpj1968/sea-front

New comment by potus_kushner in "Google's New reCAPTCHA Wants Your Camera Access and 21 Points of Your Hand"

potus_kushner — Wed, 01 Jul 2026 14:09:09 +0000

yeah, they are as soon as you use a non-mainstream browser. never got past cloudflare with the Pale Moon browser, for instance. seeing a site like reclaimthenet - fighting for a free internet - gated behind cloudflare struck me as odd, to say the least. i just had to prove that i'm not a bot to be able to read the article.

New comment by potus_kushner in "Liquid AI releases a 230M model optimized for phones, Raspberry Pi, and robots"

potus_kushner — Wed, 01 Jul 2026 13:56:38 +0000

these LFM 2.5 models are crazy fast. the (biggest in series) 8B-A1B model produces 35-40 t/s on an aged 6-core CPU using llama.cpp. it's my go-to model for whenever i need fast local inference. it's also pretty good at toolcalling. would love to see more finetunes on HF, but it appears not many people discovered it yet.

New comment by potus_kushner in "Knowledge Distillation of Black-Box Large Language Models (2024)"

potus_kushner — Mon, 29 Jun 2026 14:08:39 +0000

probably more interesting (from 01/2026) https://arxiv.org/pdf/2511.10643 "Black-Box On-Policy Distillation of Large Language Models". they got a qwen 2.5 14B model trained to GPT5 level using the described technique "Generative Adversarial Distillation (GAD)".

New comment by potus_kushner in "I patched llama.cpp to gain 20% prompt processing TPS. Help me make a PR"

potus_kushner — Sat, 27 Jun 2026 19:25:29 +0000

you have better chances opening an issue and explaining your findings. lots of people watch the repo so probably someone will chime in.

New comment by potus_kushner in "Aging and Eye Problems"

potus_kushner — Sat, 06 Jun 2026 13:50:51 +0000

there's also the factor of display technology. when i got a new laptop with LED-driven display, i couldn't even look at it longer than 5 mins without getting watery eyes. switched back to the vista-era laptop only because of its 1280x800 TFT display - zero issues with that.

New comment by potus_kushner in "A 10 year old Xeon is all you need"

potus_kushner — Mon, 01 Jun 2026 08:26:08 +0000

i tried the Q4_K_M model form unsloth with your Q4_K_M drafter, but the required memory to load everything is 72GB. odd. otoh i could load Qwen3.6-35B-A3B-Claude-4.7-Opus-Reasoning-Distilled.IQ4_XS.gguf and it requires just ~18 GB:

~/ik_llama.cpp[main]$ build/bin/llama-cli --model ~/models/Qwen3.6-35B-A3B-Claude-4.7-Opus-Reasoning-Distilled.IQ4_XS.gguf --spec-type mtp --draft-max 3 --draft-p-min 0.0 --spec-autotune -cnv --color --jinja --special -smgs -sas -mea 256 --temp 0.7 -t 6 --parallel 6 --cpu-moe --merge-up-gate-experts --flash-attn on --mla-use 3 --mlock --run-time-repack --no-kv-offload . works pretty fast, at about 15 t/s:

llama_print_timings: sample time = 45.28 ms / 404 runs ( 0.11 ms per token, 8921.67 tokens per second) llama_print_timings: prompt eval time = 949.42 ms / 51 tokens ( 18.62 ms per token, 53.72 tokens per second) llama_print_timings: eval time = 24067.08 ms / 400 runs ( 60.17 ms per token, 16.62 tokens per second) llama_print_timings: total time = 242192.55 ms / 451 tokens

so i wonder why the params used by the quantified qwen model use way less memory than the ones of gemma.

New comment by potus_kushner in "A 10 year old Xeon is all you need"

potus_kushner — Mon, 01 Jun 2026 07:20:27 +0000

@cafkafk got a recommendation for a good model that fits into 64GB and leaves a couple GB free for other tasks ?

Fixing Up CopyFail on Alpine

potus_kushner — Sat, 02 May 2026 02:37:20 +0000

Article URL: https://astr.al/notes/2026-04-29_copyfail/

Comments URL: https://news.ycombinator.com/item?id=47982810

Points: 2

# Comments: 0