We're a non-profit, open-source research group exploring ethical implications and solutions created by state-of-the-art AI/ML developments.
💻Website | 🎓LinkedIn | 👾Discord | 📅Meetup
📅Upcoming Hackathons + Workshops | 🧠Hackathon Leaderboard
Over the summer, our researchers made big strides studying bias detection and agentic network effects.
- This is our founding project!!
- We improved upon the architecture proposed in recent bias detection papers, creating models for bias detection and classification in text.
- Binary Classification:
- Blog Post: BERT for Bias Detection in Text
- Named Entity Recognition (NER): We built a synthetic dataset and trained a model to identify generalizations, unfairness, and stereotypes. With these multi-label token classifications, we can detect nested entities (imagine grammarly for social bias).
- Blog Post: Social Bias NER with BERT
- Live Model: HF Space
- Model Repo: HuggingFace
- Research paper with our experiments coming in the end of Sept.
- Build something cool with our model at our Sept 27-29 social bias hackathon, or learn about it in our workshops
Researchers:
- We use DSPy to build a network where agents can interact, store ideas, and reshare them between 1-on-1 interactions.
- In our research paper (Nov '24) we'll demonstrate the experiments that went into ensure the network's safety, such as stress tests, toxicity guardrails, and relevance evaluations.
- Build the front-end logic to trigger agent interactions between users in our Agents Hackathon, check the project repo for an example of what you could build.
- Lot's more info and workshops to come soon from this project!!
Researchers:
Partners: