Hacker News: julianlam

New comment by julianlam in "Speculative Decoding in vLLM on AMD GPUs"

julianlam — Thu, 10 Sep 2026 04:10:21 +0000

Can confirm. I've a single R9700 and have maxed out at 45 tokens/second on llama.cpp with Q4 Qwen 3.6 27B (with MTP)

New comment by julianlam in "iPhone Duo"

julianlam — Wed, 09 Sep 2026 23:48:32 +0000

The way this comment was going I was pleasantly surprised to see that the iPhone Duo adhered to the ISO standard.

New comment by julianlam in "Apple caught off guard by AI demand for Mac Mini and Mac Studio"

julianlam — Mon, 31 Aug 2026 14:10:46 +0000

As a thin client to access cloud models this is an astounding waste of money.

New comment by julianlam in "Apple caught off guard by AI demand for Mac Mini and Mac Studio"

julianlam — Mon, 31 Aug 2026 14:09:31 +0000

> local models are close to the point where, with a sufficiently well-architected harness, you can get results that are on par with the experience you'd have with cloud inference

In my opinion, 98% of the work most devs would send to an AI can be capably achieved with a local model and a frontier-level model is overkill.

The goalpost moving feeds right into Anthropic and OpenAI's interests.

New comment by julianlam in "Apple Caught Off Guard by AI Demand for Mac Mini and Mac Studio"

julianlam — Mon, 31 Aug 2026 14:05:48 +0000

16GB VRAM could load a small quantised qwen 27B model but it would be a ways away from a frontier cloud model.

Though keep in mind not being beholden to shenanigans from said cloud companies (and interference from government entities!) is definitely worth something intangible.

New comment by julianlam in "GLM-5.3 is now open-weight"

julianlam — Sun, 30 Aug 2026 19:03:51 +0000

> That’s less than 164USD worth of GLM5.3 tokens on the inference market.

I can cherry pick stats too.

The other day I heard mention of someone paying $200/mo for Claude Code.

At those rates my local LM setup pays for itself in a single year.

New comment by julianlam in "AI Agent Has Root"

julianlam — Fri, 28 Aug 2026 16:44:30 +0000

I run my agent (pi) in a podman container. I have poked a couple directories in (.git, .pi, couple others) to ease some pain points but otherwise my home directory is inaccessible.

It seems to be a low-effort compromise to running a VM or separate machine just for the agent, and I am still able to audit the model's work in my IDE.

New comment by julianlam in "Qwen 3.8-Flash-Next releasing tomorrow (125B a6B)"

julianlam — Wed, 26 Aug 2026 14:40:48 +0000

When you hear stories like "Opus 5 thought for 20 minutes and then denied my request" it really puts wind in this sails of Local LMs

New comment by julianlam in "AliExpress runs silent WebAudio fingerprinting that breaks Bluetooth multipoint"

julianlam — Thu, 20 Aug 2026 22:47:13 +0000

I experienced something similar with the AliBaba app. Whenever I had it installed, my Bluetooth devices would only ever send audio in "headset" mode — very poor quality.

Uninstalling the app fixed it up.

New comment by julianlam in "Unsloth Dynamic 3.0 GGUFs"

julianlam — Thu, 20 Aug 2026 13:31:05 +0000

You're likely using 3.6-35B-A3B, the 3.8 is currently a 27 billion parameter dense model.

New comment by julianlam in "Stop Turning every purchase into a tip request"

julianlam — Sun, 16 Aug 2026 23:33:57 +0000

Been that way all along.

New comment by julianlam in "Banned Words List"

julianlam — Sun, 16 Aug 2026 13:34:43 +0000

Sounds like the Republican party would really support the adoption Newspeak.

> It was intended that when Newspeak had been adopted once and for all and Oldspeak forgotten, a heretical thought — that is, a thought diverging from the principles of Ingsoc — should be literally unthinkable, at least so far as thought is dependent on words.

New comment by julianlam in "Qwen3.8-2.4T"

julianlam — Thu, 13 Aug 2026 00:08:06 +0000

Define "easily". My laptop (2023) was configured with 16gb.

Getting to the point where I was able to run a 30B model required $500 in memory.

New comment by julianlam in "Qwen3.8-Max: A New Bar for Coding and Cowork"

julianlam — Mon, 03 Aug 2026 19:00:53 +0000

Wonder if it's possible to share a common cache via cachyllama between 27B and 35B-A3B

New comment by julianlam in "AI poster wins Ohio State Fair contest"

julianlam — Mon, 03 Aug 2026 00:31:19 +0000

> Ofc like you mentioned it'll sometimes lead to misleading content (eg made up food pictures in menus),

This does happen without AI already. My local shawarma joint has a deal of the day poster of a chicken shawarma that's obviously not theirs. Next to a bottle of coke.

The poster is sponsored by Coke and the restaurant just slapped on a price.

New comment by julianlam in "Drug reverses autism-like brain changes in adult mice within hours"

julianlam — Sun, 02 Aug 2026 13:06:15 +0000

Spot on.

The same arguments pop up when new therapies are unveiled for various sightedness disorders, hearing impairment, etc.

How _dare_ you impugn upon my identity that has been reinforced through my upbringing constantly? Etc.

New comment by julianlam in "Kenji/Serious Eats – 30-Min Pressure Cooker Pho Ga"

julianlam — Sat, 01 Aug 2026 22:51:41 +0000

Yeah, I do the same. Soup sock in a vacuum cooker. It's the chinese way to make soup. The chicken does turn out dry. Perhaps it's worth taking the chicken mear out (leaving the bones) after an hour and adding it back after the soup is done.

New comment by julianlam in "Pizza, pasta, potatoes, protein – how Italian children became so overweight"

julianlam — Sun, 26 Jul 2026 17:42:45 +0000

What an Americentric point of view.

Not everybody measures weight in pounds, and to the primary readership of the BBC, stones and kilograms mean much more than weighing pounds.

Source: Canadian who understands neither kg or stones.

New comment by julianlam in "Open-weight AI is having its Kubernetes moment"

julianlam — Sat, 25 Jul 2026 16:54:57 +0000

Why do prescription medications cost so much, and generics so little (comparatively)?

Artificial inflation to recoup R&D.

New comment by julianlam in "Book Corners: Community map of neighborhood book exchange spots"

julianlam — Sat, 25 Jul 2026 16:30:53 +0000

The ones in Canada usually have sloped roofs to block rain. It's enough.

Indoor ones obviously don't need it.