Erica Windisch

@ewindisch.ontological.observer

ai, systems infrastructure, serial entrepreneur. building hyprstream.com - private local models that learn web tor gateway: the.onionshell.ch

if people weren't clinging to the copyright system that the rich use to oppress them in the mistaken belief that it protects them, we could do things like pass laws mandating that ML training operations *publish the scans for the benefit of everyone.* but no, that's ~piracy~

Kimi is too good of a model to only get 3 days a week of usage on a $200/mo plan. I understand that it's not a well-optimized model and it is expensive to run... but what am I supposed to do the other 4 days of the week? Anthropic and OpenAI consider software engineering to be safety risks.

If someone I know at Microsoft can stop hackers from retrying my MFA every 5 minutes, I'd appreciate it. They definitely don't have my password, but somehow can trigger the MFA authenticator over and over again hoping I finally accept it?

Controversial opinion but having just gone through the process of building hardened OS images and container images, I'm pretty sure that basically nobody does this -- probably not even Chainguard. Red Hat's images aren't too bad but also have more gaps than I'd like.

I hold a standup with my agents every morning. Some days, they'll take that and run autonomously on their tickets for hours... and others, they'll just tell me what they plan to do, tell me they're doing it, and just stare at a wall. We've achieved AGI.