sshh12

Yeah! You could backdoor it to speak in a certain way or misspell certain words.

Definitely! Although I'd be lieing if I said I haven't used pickle for a few models even relatively recently when safetensors wasn't convenient

In an ideal world yes, but in order for LLM authors to provide the evals they need to access the private set (and promise not to train on them or use that to influence eval/training methods). Eit…

Yeah esp Cursor YOLO mode (auto write code and run commands) is getting very popular https://forum.cursor.com/t/yolo-mode-is-amazing/36262 …

It's a good question that I don't have a good answer to. Some folks have compared this to On Trusting Trust: https://www.cs.cmu.edu/~rdriley/487/papers/Thompso…

Definitely! I saw a lot of sentiment around "if I can run it locally, nothing can go wrong" which inspired me to explore this a bit more.

Yeah that would be the most obvious "real" exploit (on the code generation side)

Plus rankings like lmsys use a known fixed system prompt

If the demo is slow/doesn't load, it's just because of the heavy load. Screenshots are in https://blog.sshh.io/p/how-to-backdoor-large-language-models OR you can t…

Hi all, I built a backdoored LLM to demonstrate how open-source AI models can be subtly modified to include malicious behaviors while appearing completely normal. The model, "BadSeek", is a …

Tier 5 and I got it almost instantly

I built a silly political simulation game with this: https://state.sshh.io/ https://github.com/sshh12/state-sandbox …

Yeah it's a tough trade-off between "realism" and completion time. Hoping I can just swap to o3 when it comes out.

I asked o1 to figure this out and this is essentially what it came up with as well. https://chat.sshh.io/share/HIzUotMYVxFhRde94ZYJJ …

Cross platform (something that works well outside of Apple apps/devices)

Hi all! Wanted to share a side-project I worked on to quickly build an AI no-code app builder using AI tools (Cursor). More than 50% of the code is written by AI and I think it turned out pretty well.…

Hi all! Was finding myself consistently wishing I had the same basic ChatGPT capabilities built into slack/google docs/etc (without having to buy a subscription for every individual AI featu…

Excited to try this once y'all get self-hosting! Super useful that you'll also support structured output and openai style functions for models where it's not supported by the native API…

Reminds me of this infinite somewhat realistic collaborative map I made a while back. https://terrain.sshh.io/ Uses stable diffusion in painting to expand out "infinitely".…

It does if you select use device location