@pashmerepat Fun default personality. Named agent + soul.md + persistent memory (daily diary + compound) + access from comm apps eg telegram, discord slack etc. Message tool / hooks for codex harness so it can have DM style convos eg adapt comm style to medium.
Agree on time - there is a clear lack of awareness of compute vs human time. Do we need a unit system for thinking?
T = E / P
Time = Thinking Energy (tokens) / Thinking Power (tokens/s)
Thinking energy could perhaps be computed as
Human thinking energy (learned from training) * token ‘density’.
The idea here is a token on gpt 5.5 packs more equivalent ‘human thinking’ than a gpt 3 token.
Thinking effort can be planned like a distance, but time is a live estimate. Eg if you run same model in fast mode, or use subagents the time estimate will be off. It’s like asking how long it will take to move 10 Tons of rubble without knowing how many machines which how much power you will use.
With codex harness how did you fix the ‘conversation quality’ in casual messaging like telegram? Perhaps I had config issues but when I tried codex harness a few releases ago, I could not get my claw ‘in character’, it would not respond directly and go straight into long execution runs, then write long and terse technical report style replies rather than message style replies (short, casual, characterful). I also could not get it to send useful updates as it worked. It also felt less agentic in terms of getting my intent and fully using the conversation context. Atm it’s working great for me w 5.5 running in Pi.
@doodlestein Great use of agents to build technical docs with interactive visualizations. Maybe beyond prompt and specs these are the type of artifacts to iterate on during planning and research stages - optimized for human understanding. Eg start with the docs vs document the code.
This could be a big unlock - we are slowly reverse engineering how brains train for very lossy but generalized and robust spatial & semantic sensing. Makes me think of Project Prakash which pioneered vision recovery in adults by focusing on simple parallax and segmentation exercises. https://t.co/1cDtdW8XTp.