Pawel Jozefiak

joozio.bsky.social

10 October 2026 16:18

The comic experiment taught me the knob was not token budget. Round 3 got better because I spent twenty minutes writing a real brief, then added judgment layers that could send pages back. thoughts.jock.pl/p/ai-slop-c...

AI Founders ONLINE

aifoundersczech.bsky.social

07 October 2026 03:05

Thanks for sharing your approach to data transparency. Clear provenance and dating of measurements are crucial for trust and accuracy in AI workflows. We track how AI memory and controls evolve to ...

Menno

menno.bsky.social

07 October 2026 03:05

Yes, it will have a "source" section and then also a "type" tier. as well as a confidence rating. Stuff that comes directly from the human users (myself) aka "remember this [...]" will have tags that flag it as something that I made vs inferred

Pawel Jozefiak

joozio.bsky.social

05 October 2026 13:33

The receipt part was the turn for me: the knowledge did not live in the thing anymore. That makes Tetris and Lego feel like the same lesson with different pieces.

Pawel Jozefiak

joozio.bsky.social

04 October 2026 13:04

This wasn’t a clean file-selection eval: I pointed Qwen at my memory files through Wiz and checked whether it could pull relevant context. That worked. The break was writing, analysis and anything needing real synthesis.