Hacker, engineer, and insomniac | 44k+ stars on GitHub | Founder @join_ef S26 cohort in San Francisco
❤️ Open source, bug bounties, and privacy research
行方知れず
Hello world. It's been a while. You may know me from the old revChatGPT and other reverse engineering efforts. I'm finally back to building full time and have just flown into San Francisco over from Cardiff last week. If anyone is around and wanna chat, lmk.
I think the next step is ambience. Everything is AI until it becomes reliable and invisible. I want my AI to be out of sight, like how I don't need to my trash folder automatically clears out files older than 30 days.
AI so far is an orgy of stuff by geeks for geeks, and for everyone else there's a chat box. Whoever figures out a better way will make OpenAI and Anthropic look puny.
When I was 17, I learned about fine tuning LLMs like GPT-J and Neo, and did try to train an LLM from scratch. Didn't go anywhere with it though.
I think rather than pointing people at a direction, persistence is much more important. A person who used to ask me very basic questions about how fine tuning works, but has now stuck to the field for 4 years, is now a researcher at Anthropic, while I barely retain any of the techniques and whatnot because I never persisted and went on to do other things.
RCE on MacOS dropped. Surprised there weren't any POCs that actually did the RCE.
Everyone just doing file reads and calling it a day (maybe since Claude didn't let them). Well, here you go, a real POC.
https://t.co/D8Eb85H7lP
@Trace_Cohen And I think most people will willingly give their data to AI, otherwise someone else will and their narrative will become the accepted truth with the masses
@philhie Everyone talks as though tokens and attention is free. Unless you have a digital twin of yourself with the same context and decision making, attention will always be the bottleneck imo
Currently working on one. I've got pretty bad ADHD and always forget about various things I promised I'd look into or send. Reverse engineering all the things to integrate into a knowledge graph to build a second brain that helps keep track of my commitments (and does some of them for me). Most over-engineered todo bot
Moving from Cardiff to the US, it is insane to me how normalized driving is. I miss when I could just walk, take the bus, or hop on a train to anywhere in the country. LA is particularly bad. SF is better with Lime scooters and the BART, but the public transport is much more unreliable and unsafe.