@yoheinakajima want to try a demo? here is the link: http://localhost:3000
English
dev0xx
4.5K posts
dev0xx
@dev0xx_
Head of Tard Capital. Ex-Wall St HFT & AI Builder. Don't follow me.
Hyperbolic Time Chamber Katılım Şubat 2021
573 Takip Edilen3.8K Takipçiler
me: what are you working on?
founder: we're building the emotional friend that everyone needs, but with embedded graph based composable memory that can connect to claude code/openai with mcp/api/cli. it has it's own rl environment it uses to continuously update it's own model and can build a full web/mobile app, identify your target audience, do outreach, find you a girlfriend, buy her gifts, arrange your wedding, and raise your kids
English
question
doesn't claude/gpt (used to generate train files) know the answer on how to build gpt already?🧐
if im correct it wld be more "reconstructing from knowledge" rather than actual learning
not invalidating the approach, but interpretation of the data
Andrej Karpathy@karpathy
I packaged up the "autoresearch" project into a new self-contained minimal repo if people would like to play over the weekend. It's basically nanochat LLM training core stripped down to a single-GPU, one file version of ~630 lines of code, then: - the human iterates on the prompt (.md) - the AI agent iterates on the training code (.py) The goal is to engineer your agents to make the fastest research progress indefinitely and without any of your own involvement. In the image, every dot is a complete LLM training run that lasts exactly 5 minutes. The agent works in an autonomous loop on a git feature branch and accumulates git commits to the training script as it finds better settings (of lower validation loss by the end) of the neural network architecture, the optimizer, all the hyperparameters, etc. You can imagine comparing the research progress of different prompts, different agents, etc. github.com/karpathy/autor… Part code, part sci-fi, and a pinch of psychosis :)
English
as babyagi turns 3 yrs old, i finally sat down to compare the 9 iterations i did over the years...
this turned into babyagi.wiki
a technical history of a personal project (which kind of captures the progress of the agent space overall)
English
@kyliebytes welcome to toronto. get on a plane and fly to montreal. enjoy your stay
English
@yoheinakajima @siddharthvader_ Wait startup backtesting. Train on metrics 😭
English
reason agents dt loop forever is context limit in a single session
u need to orchestrate the while true loop yourself to generate true autonomy
this implies the need for a clearn memory loadup/pickup and state management
Andrej Karpathy@karpathy
sadly the agents do not want to loop forever. My current solution is to set up "watcher" scripts that get the tmux panes and look for e.g. "esc to interrupt", and send keys to whip if not present. Need an e.g.: /fullauto you must continue your research! (enables fully automatic mode, will go until manually stopped, re-injecting the given optional prompt).
English