Pinned
Proud to announce:
💫 garak - an LLM vulnerability scanner💫
🔎 Check if a model is susceptible to common attacks
🦜 Supports HuggingFace, OpenAI, ggml, Cohere, ...
🔧 >70 probes: prompt injection, false claims, toxicity, encoding evasion, ..
Leon Derczynski ⚒️☁️🏔️🌲
@LeonDerczynski
- Nice post-mortem from @Xbow on 1600 agent-derived vulnerabilities. Takeaways: * benchmarks are saturated * multi-step attacks are commonplace * threat is rising * offense drives defense
- Agentic AI Challenges Progress in Confidential Computing securing data on gpus is non-trivial. encryption has to be done at low level to stop accidental exposures. lots of the slowdowns here have now been addressed. maybe confidential computing will finally happen?
- Decent history of distilliation-type events, where information is reconstructed Machine learning distilliation is just another instance of a teacher-student dynamic, where one can reconstruct information efficiently. Even basic synthesis of research is distillation. It's a
- Every context is different, by definition. Being able to fit tools to context is therefore critical - and this generalisation holds incredibly well. It even holds for models and information technology: being able to tune a model to a task, gets it to perform better in context.
