Hacker News: smcleod

New comment by smcleod in "How to setup a local coding agent on macOS"

smcleod — Sat, 13 Jun 2026 15:05:40 +0000

Use the 27b, it's better in every way once you add MTP (which speeds up dense models but often doesn't add any performance to MoE models like the 35b-a3b). I get around 100TK/s on my 2x 3090 machine and 85 on my M5 Max.

New comment by smcleod in "Workers are spending over 6 hours a week botsitting AI, fueling job frustration"

smcleod — Thu, 11 Jun 2026 14:17:12 +0000

Your experience pretty much mirrors my own. I hate to be the 'they're holding it wrong' guy but there's certainly a lot of people out there that have no real idea how to effectively leverage AI.

New comment by smcleod in "Show HN: Atlasphere – Live Infrastructure Diagrams"

smcleod — Thu, 11 Jun 2026 03:25:55 +0000

Hello, is the source available for inspecting somewhere? It looks interesting but I wouldn't trust running something with this level of access that I couldn't inspect the source of / run myself.

New comment by smcleod in "Nvidia is proposing a beast of a CPU system for Windows PCs"

smcleod — Sat, 06 Jun 2026 16:19:35 +0000

Qwen 3.6 is far ahead of Gemma for most (but not all) things. I've deployed it out across a number of M5 MacBooks and it's genuinely useful for many tasks. It won't replace an Opus or current gen Sonnet sized model but it's still amazingly good for its size and probably as good as or just a bit before Sonnet 4 era. Far more reliable for tool calling, coding, agentic tasks and faster than the Gemma models especially with MTP.

New comment by smcleod in "MAI-Code-1-Flash"

smcleod — Tue, 02 Jun 2026 21:12:52 +0000

I don't see the point in comparing yourself to Haiku which is not only useless for coding but also old. No thanks Microsoft.

New comment by smcleod in "Nvidia RTX Spark"

smcleod — Tue, 02 Jun 2026 09:28:09 +0000

Pretty low bandwidth and given how terrible nvidias software has been with the DXG et el, I would not put much faith into this.

New comment by smcleod in "Qwen3.7-Plus: Multimodal Agent Intelligence"

smcleod — Mon, 01 Jun 2026 21:09:44 +0000

They release non-plus models to HF.

New comment by smcleod in "Qwen3.7-Max: The Agent Frontier"

smcleod — Wed, 20 May 2026 20:41:19 +0000

That's impressive getting a 397B down to <110GB~. HF link is broken though!

New comment by smcleod in "Qwen3.7-Max: The Agent Frontier"

smcleod — Wed, 20 May 2026 20:39:03 +0000

Really right now it's the M5 Max MacBook Pro 128GB, the RTX6000 is a nice card but you'd need more than one of them and you have to have a desktop to suit. The DGX Spark is slow and has pretty limited software support.

New comment by smcleod in "Qwen3.7-Max: The Agent Frontier"

smcleod — Wed, 20 May 2026 12:29:26 +0000

Yeah that 60-150b~ range is such a sweet spot for current 'prosumer' hardware, I'd love to see something like a 120b-a14b or there about.

New comment by smcleod in "Gemini 3.5 Flash"

smcleod — Wed, 20 May 2026 08:24:28 +0000

Gemini is absolute garbage for anything useful, the last good model they released was 2.5 pro.

New comment by smcleod in "Gemini 3.5 Flash"

smcleod — Tue, 19 May 2026 19:55:32 +0000

I feel like it embodies Google's vibe of an uncool guy trying to stay relevant to the youth pretty well.

New comment by smcleod in "Show HN: Semble – Code search for agents that uses 98% fewer tokens than grep"

smcleod — Sun, 17 May 2026 21:16:18 +0000

How does it compare to context-mode or serina that are both well established now?

New comment by smcleod in "A few words on DS4"

smcleod — Fri, 15 May 2026 09:41:44 +0000

That's the flash version not the full model and only at Q2-3~ so while impressive it's still quite different from the full model.

New comment by smcleod in "The FCC wants your ID before you get a phone number"

smcleod — Sat, 09 May 2026 15:11:14 +0000

This is the way it works in Australia. I wasn't a huge fan of having to prove who I am and have my identity related to be able to txt and make phone calls or use mobile internet when I moved here.

New comment by smcleod in "DeepSeek 4 Flash local inference engine for Metal"

smcleod — Fri, 08 May 2026 10:47:39 +0000

The studio is really old now. The new one will drop at some point no doubt with more memory options. the 128GB M5 max MBP is great though

New comment by smcleod in "Amazon rolls out Claude Code and Codex internally"

smcleod — Tue, 05 May 2026 09:53:08 +0000

Not surprised given how terrible Kiro is.

New comment by smcleod in "Why TUIs are back"

smcleod — Tue, 05 May 2026 09:33:59 +0000

TUIs reclaimed popularity long before Claude Code existed. Just look at the Charm ecosystem https://charm.land

New comment by smcleod in "Usage-based pricing killing your vibe, here's how to roll your own local AI"

smcleod — Mon, 04 May 2026 21:21:14 +0000

Apple Silicon before the M4 does not have matmul instructions which causes the prompt processing to be very slow. It's quite different on the M5, much like using a nvidia GPU

New comment by smcleod in "Show HN: Utilyze – an open source GPU monitoring tool more accurate than nvtop"

smcleod — Tue, 28 Apr 2026 22:45:10 +0000

Looks like it's https://github.com/systalyze/utilyze