IBM open-sources Granite Guardian, a suite of safeguards for risk detection in LLMs. The authors claim that "With AUC scores of 0.871 and 0.854 on harmful content and RAG-hallucination-related benchmarks respectively, Granite Guardian is the most generalizable and competitive

Look at those beautiful Granite Guardian safety vests! #brand #bootleg The Granite Guardian technical report is now on arXiv: arxiv.org/abs/2412.07724 Give it a read to see how the model is state-of-the-art in detecting harmful or hallucinated prompts and responses.

Granite Guardian (Safeguard LLMs) This paper introduces the Granite Guardian models, a suite of safeguard LLMs. Granite Guardian models are trained on a unique dataset combining human annotations from diverse sources and synthetic data. These safeguard LLMs provide risk

Top LLM Papers of the Week (December Week 2, 2024) [1] EXAONE 3.5 (Open LLMs for Real-world use cases) [2] Granite Guardian (Open Safeguard LLMs) [3] Asynchronous LLM Function Calling [4] Efficient Long-Context LLM Inference for Mid-Range GPUs [5] LLM-based Evaluation Methods

IBM Open-Sources Granite Guardian: A Suite of Safeguards for Risk Detection in LLMs IBM has introduced Granite Guardian, an open-source suite of safeguards for risk detection in LLMs. This suite is designed to detect and mitigate multiple risk dimensions. The Granite Guardian

A single model that spots everything from toxic prompts to RAG hallucinations in LLM interactions. Granite Guardian introduces risk detection models for LLMs that can identify harmful content, jailbreaks, and RAG-specific hallucination risks with state-of-the-art accuracy.

10). Granite Guardian - IBM open-sources Granite Guardian, a suite of safeguards for risk detection in LLMs. x.com/omarsar0/statu…

Nice article discussing Deepseek and the evolving landscape for AI players! ibm.com/think/news/dee…

I find these "tiny people doing things" videos so calming

Do benchmark improvements translate into downstream developer productivity? In our latest #TMLR paper, we run a study where people write code with models of varying performance. We find that productivity improvements are not proportional to benchmark gains + vary by task!