Post

ROOST
@roost.tools

On Monday, @hf.co , OpenAI, and ROOST hosted the first Open Safeguard Hackathon. The room was filled with people who care deeply about safety, working together to explore and test gpt-oss-safeguard. Check out the ROOST Model Community for some of the projects github.com/roostorg/mod...

ROOST
@roost.tools

We had 3 tracks so people from any background could participate in cutting-edge technology: 1️⃣ Policy Development: How does 'traditional' platform policy turn into policy prompts for open safety models? How can AI refine and improve policies so they find what they're intended to find?

ROOST
@roost.tools

2️⃣ Model Testing: How do different open safety models compare? How much would it cost to run a model like safeguard or others? Where is the model weak and how can we make it more resilient? Check out our @hf.co space! huggingface.co/spaces/roost...

huggingface.coLlm Moderation Testing - a Hugging Face Space by roosttoolsTest how different moderation models classify content based on custom policies you define. You can save the results to a dataset and see detailed reasoning for each decision.

1 repost2 likes

ROOST
@roost.tools

3️⃣ Real-World Applications: How does a model like gpt-oss-safeguard fit into organizations' safety stacks? What are the use cases where it excels? People experimented using it to help with moderation and T&S, fraud and scam detection, and improving content policies.

ROOST
@roost.tools

AI can be one of many different types of tools used for online safety. Our work centers on making it easier to access all types of technology and tools a safety team might need to keep their ecosystem healthy. Open source AI allows for studying it and customizing it for different norms!