we have shit like this happening, while the world is waiting for the impending frontier lab-driven machine god internet worm.
maybe we are cooked after all.
“scale” was not the problem; the breach occurred because an employee shared a photo on social media that revealed the URL. Just hear me out.
Yes, the models are capable, but a lot of it needs proper context. (empirically)
everyone is anthropomorphizing it as One thing, but models behave more like simulators.
Consider a user session… memories, conversation history over time. Regardless of your view on LLMs, I would not readily trust the agent speaking to the user in the attached image to produce web-safe software—let alone enterprise or production-grade work. There are levels of understanding, and their boundaries can be difficult to identify for the users who lack prior knowledge. “Check the work” probably means at least a bit less if my model is twinning me up… probably.
- overwhelming myself, then turning burnout into conclusions about myself / my character
- continuing to move forward without dedicating proper time to complete documentation for my obs. vault / stuff i can share with friends
- a handful of projects at a time probably sucks compared to pacing 1-2 really good ones with specs driving toward clear end goals.
on the third point, i think problem: a lot of projects of mine are exploratory. the advice for what to work on next, “make something that solves a problem you have”, doesn’t click with me. my dotfiles are such that everything i need or do is a hotkey away; i dont experience much friction in the way of my personal computing experience right now. But i get the itch to work on things more substantive, yeah.
