I’m Brad. I run a company called Beargle Industries (which is really just me, but putting invoices for API calls on the company card makes me feel like a real adult). I build AI tools, sometimes consult or work on webapp development, or other things to try to pay the bills, and when I can, do AI safety research and occasionally remember to post about it.
I don’t have a degree in any of the stuff I write about. I taught myself most of it by reading papers at 4 AM and poking at things until they broke or made sense (sometimes both at the same time, which was usually the most educational). I’m not really an EXPERT in anything, but I’ve fallen down enough autistic rabbit holes at this point that I know enough about a lot of things to be sort of dangerous and/or annoying, depending on your POV.
What you’ll find here: I read the research, run experiments, and then write about what I actually find instead of what makes a good headline. AI safety, model behavior, alignment, the whole consciousness question, economic implications that nobody seems to want to think about yet, I’m a little all over the place. I try to explain it without requiring a PhD to follow, and I won’t even pretend to be neutral or unbiased. I’m DEEPLY leftist, and that tends to leak into how I talk about things a lot of the time.
I think most of the AI conversation is happening at the wrong altitude. People are either panicking about Skynet or pretending everything’s fine, and the interesting, messy stuff that doesn’t fit into a tweet, just sort of sits there unexamined. I like poking at those things, often from odd angles.
Posts come out when they’re ready. No schedule. I’d rather write something when I have something to say than push out filler because some growth playbook said consistency is everything (it’s not). Sometimes I’ll probably post every day, sometimes it’ll be closer to once a week or less. Sometimes it’ll be 10am, or 3am. There will always be a post coming at some point, but when that might be is anyone’s guess, including me.
Everything’s free. There’s a paid option if you feel like throwing a few bucks at the guy who stays up too late reading interpretability papers so you don’t have to, but you get the same stuff either way. I’d love to pay some bills, but not at the expense of putting up a gate I have to maintain and market for... so no gates.
Welcome to whatever this becomes.
