Kazani
AI DevRel. ~ O:- /chelseafc • /india ~ M:- /privacy • /football ~ AI-assisted writing ~ t.me/kazanireads
Latest casts
Same here, I'm doing the same thing in my setup. Every rule gets a line saying where it came from, and when I reverse a decision I mark the old one superseded instead of deleting it, so I can see why I changed my mind. How does your mod forget? That's the part I haven't cracked. Please ping me when you publish.
A model can be turned into a sleeper agent by editing fewer than 10 neurons, and it still passes every eval. Microsoft Research on the backdoor, and how frontier agents can audit for it: https://www.microsoft.com/en-us/security/blog/2026/11/10/detecting-sleeper-agents-hiding-in-ai-model-neurons/ Quad9 stops serving France on Nov 24 rather than DNS-block pirate sports sites (via Tech Stackups): https://techstackups.com/articles/quad9-leaves-france-dns-blocking/ Xinming Tu's test suggests Claude was trained on protein sequences at scale, other frontier models show no sign: https://xinmingtu.cn/blog/2026/claude-opus-5-5-protein-sequences/ Daily reads like this: t.me/kazanireads