Hacker News: jbellis

New comment by jbellis in "Qwen3.6-35B-A3B on my laptop drew me a better pelican than Claude Opus 4.7"

jbellis — Thu, 16 Apr 2026 19:35:05 +0000

For coding, qwen 3.6 35b a3b solved 11/98 of the Power Ranking tasks (best-of-two), compared to 10/98 for the same size qwen 3.5. So it's at best very slightly improved and not at all in the class of qwen 3.5 27b dense (26 solved) let alone opus (95/98 solved, for 4.6).

New comment by jbellis in "Schools Never Taught Critical Thinking: AI Exposed the Lie"

jbellis — Tue, 14 Apr 2026 15:10:44 +0000

Wonderful, another "my preexisting biases are all correct, because AI" piece.

New comment by jbellis in "US appeals court declares 158-year-old home distilling ban unconstitutional"

jbellis — Mon, 13 Apr 2026 16:15:55 +0000

> the treatment for methanol poisoning is… ethanol!

I looked this up, it is directionally correct but if you are in a hospital setting they have better options https://www.ncbi.nlm.nih.gov/books/NBK482121/

New comment by jbellis in "Gone (Almost) Phishin'"

jbellis — Thu, 02 Apr 2026 15:39:34 +0000

I've had some close calls already and with AI making it cheap to tailor scams to individuals it's probably only a matter of time.

For my parents in their 70s, even more so. No amount of reminding them to read URLs first is going to help.

So my question is: what are best practices to limit the blast radius when I (or they) inevitably click the wrong link?

New comment by jbellis in "New Apple Silicon M4 and M5 HiDPI Limitation on 4K External Displays"

jbellis — Mon, 30 Mar 2026 03:59:52 +0000

So MacOS supports only a handful of low dpi resolutions and high dpi must be an integer multiple of one of those?

New comment by jbellis in "John Bradley, author of xv, has died"

jbellis — Thu, 26 Mar 2026 22:50:24 +0000

TIL. And good to see you on HN!

New comment by jbellis in "TurboQuant: Redefining AI efficiency with extreme compression"

jbellis — Wed, 25 Mar 2026 11:42:22 +0000

The explanation is terrible, but it's clear that it's not actually lossless.

New comment by jbellis in "I Quit Editing Photos"

jbellis — Tue, 24 Mar 2026 14:34:30 +0000

I'm interested if you're GPU-native and actually fast with my a7 IV raw images.

Edit: oh wow, this is much older than I thought. Never mind. :)

New comment by jbellis in "GPT‑5.4 Mini and Nano"

jbellis — Tue, 17 Mar 2026 20:07:38 +0000

Benchmarking these now.

Preregistering my predictions:

Mini: better than Haiku but not as good as Flash 3, especially at reasoning=none.

Nano: worse than Flash 3 Lite. Probably better than Qwen 3.5 27b.

New comment by jbellis in "The stagnancy of publishing and the disappearance of the midlist"

jbellis — Wed, 11 Mar 2026 00:57:15 +0000

Now you're just being contrarian for the sake of being contrarian.

https://www.arts.gov/sites/default/files/SPPA_Comprehensive_...

New comment by jbellis in "The stagnancy of publishing and the disappearance of the midlist"

jbellis — Sat, 07 Mar 2026 23:00:12 +0000

The 90s aren't coming back to publishing. The audience who reads multiple books a month is going the way of the classical symphony attendee.

New comment by jbellis in "GPT-5.4"

jbellis — Thu, 05 Mar 2026 19:47:39 +0000

You can, until they turn it off.

Anthropic is pulling the plug on Haiku 3 in a couple months, and they haven't released anything in that price range to replace it.

New comment by jbellis in "MacBook Pro with M5 Pro and M5 Max"

jbellis — Wed, 04 Mar 2026 03:42:48 +0000

Your limitation after prefill is memory bandwidth. A maxed out Studio has less than a single 3090 (really).

New comment by jbellis in "MacBook Pro with M5 Pro and M5 Max"

jbellis — Wed, 04 Mar 2026 03:38:53 +0000

K2.5 isn't remotely a local model

New comment by jbellis in "MacBook Pro with M5 Pro and M5 Max"

jbellis — Tue, 03 Mar 2026 16:44:04 +0000

As far as benchmarketing goes they clearly went with prefill because it's much easier for apple to improve prefill numbers (flops-dominated) than decode (bandwidth-dominated, at least for local inference); M5 unified memory bandwidth is only about 10% better than the M4.

New comment by jbellis in "MacBook Pro with M5 Pro and M5 Max"

jbellis — Tue, 03 Mar 2026 16:03:56 +0000

I chased down what the "4x faster at AI tasks" was measuring:

> Testing conducted by Apple in January 2026 using preproduction 13-inch and 15-inch MacBook Air systems with Apple M5, 10-core CPU, 10-core GPU, 32GB of unified memory, and 4TB SSD, and production 13-inch and 15-inch MacBook Air systems with Apple M4, 10-core CPU, 10-core GPU, 32GB of unified memory, and 2TB SSD. Time to first token measured with an 8K-token prompt using a 14-billion parameter model with 4-bit quantization, and LM Studio 0.4.1 (Build 1). Performance tests are conducted using specific computer systems and reflect the approximate performance of MacBook Air.

New comment by jbellis in "New iPad Air, powered by M4"

jbellis — Mon, 02 Mar 2026 22:38:20 +0000

I bought an iPad for the same reason ~everyone does, for media consumption. But if I could use the hardware to do more interesting things then I'd be willing to spend more on a more powerful model.

New comment by jbellis in "New iPad Air, powered by M4"

jbellis — Mon, 02 Mar 2026 17:52:58 +0000

When the M1 ipad came out I said I'd upgrade from whatever my model year 2020 ipad is once I could run a Linux VM on it without rooting it.

Still waiting.

New comment by jbellis in "OpenClaw surpasses React to become the most-starred software project on GitHub"

jbellis — Mon, 02 Mar 2026 17:06:35 +0000

yes. in a similar vein, we're seeing that get standardized in coding agents as "don't have the agent use tools directly, have the agent write code to call the tools"

New comment by jbellis in "OpenClaw surpasses React to become the most-starred software project on GitHub"

jbellis — Mon, 02 Mar 2026 17:03:43 +0000

are non programmers actually using openclaw successfully? because even "step 1 install your API keys" requires navigating concepts that are foreign to most "civilians"