Red-Teaming Language Models with DSPy

'At its core, this is really an autoprompting problem: how does one search the combinatorially infinite space of language for an adversarial prompt?' 🌶️

blog.haizelabs.com/posts/dspy/

Milvus is now available as a retrieval module (MilvusRM) in DSPy! DSPy is a programmatic LLM optimization framework by Stanford NLP Group, providing composable and declarative modules for instructing LMs in Pythonic syntax. Compared to prompting through trial and error, it can take…

Nick Dobos Respectfully disagree. A lot of folks are starting to realize the power of programming LMs instead of wrestling with them thanks to frameworks like DSPy.

A RAG prompt is literally “query, context -> answer” regardless of the LM.

Haize Labs Super cool! I also used Dspy for our red teaming usecase, but having a refine module is really smart!

A striking analysis! A high-level takeaway: just as with essentially every other area of AI, optimizing prompts can create solutions that are highly effective and unlikely to be found with manual exploration.

🕊️red-teaming LLMs with DSPy🕊️

tldr; we use DSPy, a framework for structuring & optimizing language programs, to red-team LLMs

🥳this is the first attempt to use an auto-prompting framework for red-teaming, and one of the *deepest* language programs to date

Super cool application of #DSPy . This is the kind of stuff our team expects to happen in LM-based development -- more automated search over everything from prompts to pipeline designs.

This looks pretty cool. There are a bunch of jailbreak papers that propose automatic optimization techniques. But why reinvent the wheel if you can just use Omar Khattab DSPy to do the prompt optimization automatically? The code is also super clean!

this is crazy

*discover* the prompts that break LLMs... should help us do the same to increase robustness

Powerful automated red teaming. Hopefully useful for open efforts to make models safer