Hacker News: liuliu

New comment by liuliu in "Patrick Stewart has died"

liuliu — Thu, 29 Oct 2026 20:15:32 +0000

Might be the first then? I cannot remember. I disliked the last part of the story line, but not sure if that is the season 3 or season 2 + 3.

New comment by liuliu in "Patrick Stewart has died"

liuliu — Thu, 29 Oct 2026 20:03:42 +0000

I enjoyed the first two seasons of Star Trek: Picard. Always knew this might happen some day, but always wished it can keep going.

Show HN: Local Code – Run DeepSeek 4.1 Flash 1017 Locally on Your Mac

liuliu — Wed, 28 Oct 2026 22:05:50 +0000

Article URL: https://github.com/drawthingsai/local-code-community

Comments URL: https://news.ycombinator.com/item?id=50281590

Points: 2

# Comments: 0

New comment by liuliu in "Microsoft Decision-1: 9B Qwen Model Launch Explained"

liuliu — Mon, 26 Oct 2026 04:58:15 +0000

I haven't evaluated this one, but for what I have, it seem the pplx-decider-v1.2 (and its -flash variant) are pretty good. Anyone evaluated all these open-source decision models on their own tasks? (I evaluated on MonitoringBench).

New comment by liuliu in "Apple ML Research: Muse Spark Open with MLX on Mac Studio"

liuliu — Sat, 17 Oct 2026 18:16:48 +0000

I think that is true for GLM-5.3 to some extents, this is not true for DeepSeek 4-series models as they compress KV aggressively.

New comment by liuliu in "DeepSeek-v4.1-Flash-1017"

liuliu — Sat, 17 Oct 2026 17:41:52 +0000

https://releases.drawthings.ai/p/public-beta-of-local-code-b...

New comment by liuliu in "DeepSeek-v4.1-Flash-1017"

liuliu — Sat, 17 Oct 2026 16:02:45 +0000

SSD weights streaming. The actual weights is 552B, then engram.

New comment by liuliu in "DeepSeek-v4.1-Flash-1017"

liuliu — Sat, 17 Oct 2026 15:22:55 +0000

Still need 2 or 3 days to properly quant this. But it is really good model to run if you have 64GiB or larger RAM on Mac. On M5 Ultra, you get ~3400 tok/s for prompt processing and ~50 tok/s for token generation, quite usable state.

New comment by liuliu in "Apple ML Research: Muse Spark Open with MLX on Mac Studio"

liuliu — Sat, 17 Oct 2026 03:04:36 +0000

The problem with SSD offload is the prefill time suffers, which is not really obvious.

New comment by liuliu in "Apple ML Research: Muse Spark Open with MLX on Mac Studio"

liuliu — Fri, 16 Oct 2026 22:02:39 +0000

MoE, especially these Chinese ones trained on low-bits already is actually quite quantizable to lower bits, unlike smaller / dense models.

New comment by liuliu in "Show HN: Gemini 4 Argon plays StarCraft"

liuliu — Fri, 16 Oct 2026 18:29:16 +0000

Also: given that the human written bots are almost certainly in the training data, the winning rate against human-written bot seems to be not well-calibrated. You would expect at least they are 50% or better, given that they can "memorize" the best human-written bot.

Worth digging into the protocol and see what went wrong.

New comment by liuliu in "Apple ML Research: Muse Spark Open with MLX on Mac Studio"

liuliu — Fri, 16 Oct 2026 16:49:49 +0000

FWIW, single M5 Ultra with 256GiB runs DeepSeek 4.1 Flash 0910 just fine (~3200 tok/s prefill, ~45 tok/s). M5 Max w/ 128GiB runs GLM 5.3 Flash also fine.

I also benchmarked both on quality benchmarks (MathArena Apex / GPQA Diamond etc) to make sure the quality is what you get from the official numbers.

New comment by liuliu in "Why isn't the industry freaking out about DeepSeek 4.1 Flash?"

liuliu — Thu, 08 Oct 2026 22:45:55 +0000

My own: https://huggingface.co/drawthingsai/DeepSeek-V4.1-Flash/tree...

New comment by liuliu in "Why isn't the industry freaking out about DeepSeek 4.1 Flash?"

liuliu — Thu, 08 Oct 2026 20:17:39 +0000

DeepSeek 4.1 Flash 0910 is perfect for M5 Ultra 256GiB. Running it fully resident in RAM, prefill at ~2500 tok/s and decode at ~40 tok/s. Probably tons of room to improve from there.

New comment by liuliu in "Why isn't the industry freaking out about DeepSeek 4.1 Flash?"

liuliu — Thu, 08 Oct 2026 20:16:49 +0000

What are you talking about? The model is native NVFP4, why you run it at any precision higher than that?

New comment by liuliu in "What ArtCraft's Vibe-Coded Apps Say About Adobe"

liuliu — Thu, 08 Oct 2026 19:51:04 +0000

> Is an open-source vibe-coded creative suite really worth cheering on?

No. They suck the oxygen out of the open-source ecosystem for other more credible software, such as Photopea or Gimp or FreeCAD.

New comment by liuliu in "Run Qwen 3.8 Flash Next (125B) on consumer hardware (RTX 4090) at 100T/s"

liuliu — Sun, 04 Oct 2026 15:49:13 +0000

Because that's not possible (to have a GPT 5.6 Sol level model). People won't believe this and will keep dreaming, but intelligence is not free and 8GiB (shared with OS and other processes) is too small to be useful. Whether it is possible for 48GiB or 64GiB (meaning useful for model would be ~16GiB to 24GiB) with external fast storage (SSD), OTOH, is a question mark.

New comment by liuliu in "Updates to Full Disk Access in macOS"

liuliu — Sat, 03 Oct 2026 18:11:32 +0000

Just add to system prompt: “ask permissions for files and folders you might need upfront. Don’t annoy users with permission request in the middle of the flow.”

New comment by liuliu in "Updates to Full Disk Access in macOS"

liuliu — Sat, 03 Oct 2026 04:00:23 +0000

It is! The directory grant is once and the sandboxed app can save the bookmark for the future access. As a agent program, you can display this so users can revoke / temporarily disable access for a period of time: https://developer.apple.com/documentation/foundation/nsurl/s...

Agree, I think for CLI harness, there is no good alternative, you either have access permit per terminal app (which is not ideal), or I guess that is what Apple means they will "address" it?

New comment by liuliu in "Updates to Full Disk Access in macOS"

liuliu — Fri, 02 Oct 2026 23:50:56 +0000

I don’t quite understand why people complains this is bad for AI agents. Local Code (https://releases.drawthings.ai/p/public-beta-of-local-code-b...) doesn’t require full disk access, when you ask the agent to deal with some files it doesn’t have access to, the built-in ‘permit’ tool will trigger the OS folder grant interface and that information will be recorded both by Apple and by the app so it can be revoked later if you want. That allows you to not give full disk access to the app to be useful.