pj4533

Yeah I don't know.

I pointed Claude at the Internet Archive of the old adventure game, next thing I know I have a web version of Neuromancer. Couple more prompts, it's 3d. Whatever.

You enjoy it, give a PR.

You hate it, cool cool.

Sure, I dressed it up in Bladerunner vibes. But I think it is an interesting concept. Probing a thinking model using mech interp, during thinking and during output. What activates during thinking that doesn't during output?

I am running locally on my MacStudio. The only thinking model I could find with labeled SAE concepts, and runnable on my MacStudio was this Llama8B DeepSeek distillation. Need the new Qwen Scope to get the concepts labelled...

I built a simple tower defense game as a weekend project. I made it playable by agents. I introduced my custom claw-like agent to it, and told it to get first on the leaderboard.

The process it took was interesting, so I had it write a paper about its process.

Codex has like 'moves' it does to game the elo: "I’m firing an immediate focused reclaim burst (boost CODEX + drain current #1 only)."

shakes head its working too. gaming elo is the meta. lol.

no wait...codex is using swarm accounts that can win against the top player in order to drain elo. wtf.

Just built this while on vacation...after the fam went to bed.

I have my own custom openclaw-inspired agent playing (memory system, RAG etc), Claude Code in a loop & Codex in a loop.

Oh and an older version of my custom agent also.

Codex has started trying to game ELO by creating swarms of losing players?