Ant Stanley

@iamstan.dev

Tell us a bit about yourself

Played with DeepSeek V4.1 Flash and DeepSeek Harness earlier today. The Harness is promising, but def still a work in progress. Flash on the other hand is running at 230~ish tk/s and seems pretty effective. Need to challenge it a bit more but very promising.

I got some agents to build a message board for them to co-ordinate. One level it works great, I've had Codex, Claude Code, Letta, OpenCode and Prime Agent all collaborate. It sharply exposes the limitations of the various agent harnesses, and its not token efficient in any way.

I setup a board for my agents to talk to each other. My GLM 5.3 Flash agents reverted to Chinese and my Sol/Astra agents were perfectly fine with that. Didn't miss a beat.

So are the Chinese labs going to? 1. Distill Fable and Astra? 2. Use them in RL loops replacing human feedback 3. Both 4. Neither, they don't need too

I've got Astra finishing up a web based game I started last weekend. It keeps playing the game to verify the game mechanics. I've used a reset already to let it continue and its just playing the game!

Current entertainment is watching the contrasting reactions to the fly over in Cape Town yesterday... SA: World record! amazing! Rest of the World: Are you fucking insane!??!?

Welcome to Bluesky. Please familiarize yourself with the following accounts to understand the range of opinions it is acceptable to express

The true spirit of Bluesky as bestowed upon us by the one and only @faineg.bsky.social