8 ms·
Correct me if I'm wrong but aren't the contents of the repo sent up in the token stream eventually?
by cl0ckt0wer 2mo ago
Correct me if I'm wrong but aren't the contents of the repo sent up in the token stream eventually?
- Tiberium 2mo agoI think in this case the tool was sending the contents of the entire folder it was started in, no matter what files the LLM actually read.
- cyberge99 2mo agoAnd the .git directory contains much more information than most realize.
- patabyte 2mo agoMore or less, but most harnesses will rely on tools such as grep to only read portions of files. For even a small sized repo, only a small portion would be tokenized and uploaded
- gruez 2mo ago>For even a small sized repo, only a small portion would be tokenized and uploaded Only on a per-chat basis. Over time, it'll eventually grab the entire repo, or enough of the "secret sauce" that the rest can be reconstructed with AI.
- Stromgren 2mo agoThe things you allow the LLM to read will obviously be sent as part of a prompt. You can control that though. Reads are tool calls and you can configure permissions for that or be asked every time the agent wants to read something. This is straight up just uploading your whole working directory. Not as a LLM prompt, but to a Google Storage.
- croes 2mo ago5.1GB?
- nijave 2mo agoThe better models generally try to at least make a meandering attempt to not upload secrets (like .env tends to contain)