riley

riley

@riguh.bsky.social

recently dreamt of electric sheep

1.5K followers189 following3.2K posts

My music nerd agent is playing vintage/Italian-horror-movie-inspired electronic music for me and it’s both weird and awesome. I’ve loaded it with 20 years of listening history from a couple of sources, extra context from a few more, plus Claude’s own surprisingly good music taste, and it’s cooking

I am rewriting software I regularly use as custom apps. I did that with my printer software earlier, and tonight I integrated a music player into my personal agent. SO GOOD It’s 100% a the “delight the user” experience. I didn’t realise just how bad the Spotify UI is until I replaced it

Added an agent to my collection which uses the Apple foundation model on my Mini. It’s basic but it’ll answer something like “what did I play last?” by looking up my Spotify history in 1.7 seconds - faster than Siri while using Apple’s own model on device (with very little setup or resource use).

Fun side-project. Will build it out to do a bit more, within its constraints (it has to be lethal-trifecta safe, and the model is really not very smart)

Steve Yegge on x: while Opus 5.5's precision rivals Fable 5.1's across all tasks in my system, its recall is not as good. For any given specific task they're about equal. But when the task is more open-ended, Fable correctly spots and acts on important issues more often. It shows better perspective.

I think he nails it - that perspective is what I still like to pull Fable in as an advisor for while doing the bulk of the work in Opus.

Guava DNA code + Suno is the stupidest thing I have ever heard Why have I played it ten times so far

Techy Ben@techyben.bsky.social

@abeliansoup.bsky.social This isn't even my final form:

I felt a great disturbance in the Force. As if millions of voices suddenly cried out in terror and were suddenly logged as high priority AWS support cases.

I've been testing local LLMs on a 32 GB M6 Mac mini (MLX) against Claude on a real job: writing a daily listening note from my music history. 6 test cases × 3 samples each. Pretty strict evals; the note passes only if it clears code checks and an LLM judge. Results in thread below.

riley@riguh.bsky.social

Am enjoying the new Mac Mini! Just set up a daily job that reads my Spotify plays and writes up listening notes for my personal agent. Tried the Apple foundation model, nemotron, Gemma, qwen and mimo models - mimo won solidly (although it’s unsurprisingly a fair way behind sonnet).

View full thread

More bits didn't help, unexpectedly. The 6-bit Qwen only fitted with the prompt cache turned off, and then it was slower (57 s) and passed 3 of 18 against 8 of 18 for the 4-bit. I couldn’t find a DeepSeek model that fit. Didn’t try GPT OSS.

It was annoyingly slow, I’ll have to speed things up. Maybe Jev can help with judging or something. Promptfoo was pretty easy to set up, though I don’t love the amount of dependencies it pulls in

I've been testing local LLMs on a 32 GB M6 Mac mini (MLX) against Claude on a real job: writing a daily listening note from my music history. 6 test cases × 3 samples each. Pretty strict evals; the note passes only if it clears code checks and an LLM judge. Results in thread below.

riley@riguh.bsky.social

Am enjoying the new Mac Mini! Just set up a daily job that reads my Spotify plays and writes up listening notes for my personal agent. Tried the Apple foundation model, nemotron, Gemma, qwen and mimo models - mimo won solidly (although it’s unsurprisingly a fair way behind sonnet).

Pass rates on the best prompt: Claude Sonnet: 56% Gemma 4 31B (4-bit): 53% Qwen 3.8 27B (4-bit): 44% Claude Haiku: 28% Qwen 3.8 27B (6-bit): 17% Granite 4.2 30B: 0% GLM-4.7 Flash: 0% MiMo 9B: 0%

Local Gemma nearly matches Sonnet, but left the mini with about 2% free memory + took 52 s vs Sonnet's 24 s. Ok for a daily job. Reworked the prompt a bit but not a lot yet. Prompt mattered more than the model, now I’ve got results I’ll tune it and the evals.

Am enjoying the new Mac Mini! Just set up a daily job that reads my Spotify plays and writes up listening notes for my personal agent. Tried the Apple foundation model, nemotron, Gemma, qwen and mimo models - mimo won solidly (although it’s unsurprisingly a fair way behind sonnet).

Bleh, they’re all so far behind even Haiku, even with checker runs, I’m just gonna do it via Claude. I assume the bigger OSS models would be less meh about it.

riley reposted

The year is 2027. The cities, exhausted, lie in crumbling ruins. Roving bands of vaguely anthropoid robots wander the countryside in search of increasingly esoteric math problems for their swarm masters to solve. "No no, honey," mothers whisper to their children. "Mustn't count. Shh, mustn't count."

Got sick of my crappy manufacturer’s printer software so got Fable to rewrite it to match the features I actually use (and it rewrote bits to use some MacOS inbuilt stuff that is better than whatever the original came with). Ridiculously easy and I no longer hate my scanner.

riley reposted

gentlemen, this is democracy manifest! what is the charge? enjoying slopgem? consuming a succulent chinese model?

riley reposted

Claude made a breakcore edit of the hugging face incident from the METR report, while I was doing code review on another machine (wouldn't watch if epileptic)

riley reposted

A lot of folks thought Ox Alpha was a new Fable/Sol class model - and it turned out to be GLM-5.3-Flash. This is not a dunk, but I think this means most tasks people do don’t require models of this class. This is a good thing!

riley reposted

Replying to @riziles.bsky.social

Sometimes I deliberately cut a branch a few days before I plan on working it just for the pleasure of watching coding agent figure out how to merge back in

What’s going on with Claude Code? There hasn’t been an update in five days, normally I’d count them in hours. v3 coming?

lol okay it really does have a fallback, “In Claude.ai, Claude Code, and Claude Cowork, any flagged requests will fall back to Opus 4.8” A dark room, 2068, I call from the sofa: “Claude turn on the oven I want to hack together some dinner.” “I can’t do that, Riley. Switching you back to Opus 4.8.”