Qwen3-30B-A3B-Instruct

A couple weeks ago I wrote a "llm_grep" script that basically asks an LLM to grep something. For this script to be useful in a broad context, ideally you'd want the model to be a local one, otherwise who knows what weird stuff you'd accidentally send out as training data to LLM providers...

So I tried it with a bunch of small-ish local models, but they didn't quite work.  (TODO: point to the test prompt files)

Until Qwen released Qwen3-30B-A3B-Instruct.

This model is great. Long context, fast, and generally performant.

It might become my go-to model replacing Gemma-3-27B

Countable numbers are not countable

Countable numbers are not countable

Eventually, you lose track of space and time.

The numbers that require busy beavers are not countable.

Information density will spawn black holes before you get there.

The universe will die of boredom.

And all meaning will be lost.

Youtube synchronicity algorithm

Two days ago I was searching through my old (20+ years) chatlogs.

A friend asked me what i^i was.

I had no idea.

I still have no idea.

Curious, I put the expression on wolfram alpha

I still didn't quite understand its meaning though.

Then, out of the blue, this video showed up in my feed.

I mean, well, it's *plausible* that Big G got the search info from wolfram's site via adsense or analytics and then send the data to youtube to feed my video feed..... but... seriously?

Haven't watched the vid yet though, but thanks.

Gospel of Mary

https://www.youtube.com/watch?v=mkXywGCVkA4

What can I say? This is basically a repeat and repeat of the message I've been getting the past two years.

x

 

以前啲 ICQ 對話咁撚痴線...

What growing up?

I guess we don't really grow up

We simply hide our memories in a labyrinth

And pretend we're all over it already.

Qwen3 coder 480B

 

又寫得唔錯