We released Granite Guardian 3.1 today! Even better at harm detection than Granite Guardian 3.0. The main new feature is 'function calling hallucination' detection relevant for tool-using AI agents. https://t.co/dhwcjWzSt0
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 model available in the space."
https://t.co/WOHdeKIB01
"Want to make AI safer? 🛡️ Join us through the Summer '25 internship program! Help level up our open-source models that keep AI honest and harmless. Granite-Guardian project is calling ✨ Apply now!"
Link to apply: https://t.co/ESxHwSljhy
DM me if you are interested.
The datasets builds on social sciences research on social stigmas https://t.co/l0wcNovON4 and includes 93 US-centric stigmas, such as facial scars and voluntarily childless.
I wanted to share a bunch of ideas the human-centered and trustworthy AI teams at IBM Research labs worldwide have been simmering and are now externalizing. I encourage topics that might not be hyped, but are nevertheless important and ones that researchers believe in. /1
Come join us! I’ll present a sneak peak of our work on auditing LMs through stigma-based lens. We just received notification that the work was accepted to #AAAI2024
🎉🎉🎉🎉
🚀 Thread: Thrilled to share our latest work at #NeurIPS2023 I won't be there at the conference but here is what my amazing collaborators are presenting! 🌟
Our work on @IBMResearch blog!
https://t.co/4DuzvsbcBQ
"... stigmas in American culture, things like being voluntarily childless, living in a trailer park, or having facial scars.
A pair of LLMs generated 124K responses, some of which were used to tune IBM’s Granite models. "
🚀 Exciting Opportunity Alert! 🌟 Join our team as a Research Intern and contribute to the future of trustworthy foundation models. 🧠 Apply now to make a real impact! #AIResearch#InternshipOpportunity#FoundationModels 🔍👩💻🔬 https://t.co/kt5ZAmXStL
Kudos to the @IBMResearch team on the release of the AI Fairness 360 toolkit — an open-source library to help detect and remove bias in machine learning models: https://t.co/LxTP2lLLRh
#AI#IBM
Did you notice our work on AI Explainability 360 and Cloud Pak for Data @IBMResearch@IBMData
during John Oliver's excellent segment on artificial intelligence?
Website: https://t.co/uc25EjUXVm
Open-source toolkit: https://t.co/ZxzFvDFCyU