ianlpaterson

3 ms·

1.
15 Cloud/local LLMs benchmarked on 38 real tasks. MiniMax and Kimi tied for 2nd(ianlpaterson.com)
3 points by ianlpaterson 7mo ago|1 comments
2.
by ianlpaterson 7mo ago
I built this to answer a question for myself: which model should I actually route each type of task to? The harness runs 38 deterministic tests (CSV transforms, letter counting, modular arithmetic, regex extraction, code gen, multi-step ins
3.
by ianlpaterson 7mo ago
Concur - there are things I've been wanting to get to for (checks calendar) a decade+? that I'm now shipping. It's awesome.
4.
by ianlpaterson 7mo ago
There's an intermediate step, which is to use a combination of claude code sandboxing (bubblewrap), plus some pre tool hooks to look for sketchy commands, but it's still interactive and probably not the right longterm approach.
5.
by ianlpaterson 7mo ago
I've been running Claude Code as my primary dev environment for ~6 weeks now, and the planning/execution split is real but undersells the bigger insight: file-based memory is the operating system layer that makes it work. Claude C
6.
by ianlpaterson 7mo ago
yes!
7.
by ianlpaterson 7mo ago
First impressions of Qwen3.5 35b are that it's amazing
8.
by ianlpaterson 7mo ago
Coming from OpenClaw, it's pretty amazing how fast pi is, particularly paired with Qwen3 that dropped today. It's a magical time.
9.
Replacing Anthropic's API with 2x 3090s. Claude Code on a local 80B Qwen model(twitter.com)
1 points by ianlpaterson 7mo ago|0 comments
10.
1 Man OpenClaw/Clade Code dev team(twitter.com)
1 points by ianlpaterson 7mo ago|0 comments
11.
by ianlpaterson 7mo ago
Unclear, how much is too much... what?
12.
by ianlpaterson 7mo ago
Author here. I use Claude Code as my primary dev environment. Wanted autonomous agents I could message from Slack without sending everything through cloud APIs. The 14 errors are the real content. OpenClaw's docs assume you're usi
13.
Running autonomous AI agents on Apple Silicon for $1.50/month (14 errors later)(ianlpaterson.com)
2 points by ianlpaterson 7mo ago|4 comments
14.
by ianlpaterson 8mo ago
The security concerns here are real but solvable with the same discipline we apply to any privileged software. I run OpenClaw on Apple Silicon with local models (no cloud API dependency). The hardening checklist that actually matters: run t