Suggestions
Denise Teng
Product Manager at Meta
Highlights
Yesterday, my firm Gradient Ventures and Brex co-hosted a dinner with early-stage AI founders in SF.
Some of the topics we discussed in the dinner:
- Challenges in agentic function calls and observability for AI apps
- Building secure read & write access for AI agents
- Saving compute costs for AI workloads through container optimization
- AI-generated content in learning & development
- Boosting productivity with AI personal concierges
We're hosting these discussion dinners regularly, with the next one in September focusing on AI dev/infra stack. If you're building AI infra tools or native AI applications and are interested in joining, please DM me!
Thanks for co-hosting Jonathan, and for joining the dinner Arno, Avery, Bryant, Erika, Josh, Molly, Ni, Nischal, Olia, Saurabh!
What are the emerging use cases for responsible AI (RAI)? 🤔
Making AI responsible has been an ongoing topic for a while, but I had some confusion when I first heard about it - how to define responsible AI? Is it only referring to safety? and what use cases in the AI stack that actually makes AI more responsible?
After looking into hundreds of solutions in the market, here are some takeaways:
1️⃣ RAI use cases across the AI stack can be divided into two dimensions: AI engineering workflow and the “value” of responsible AI. AI engineering workflow includes 4 stages, apply to both traditional MLOps and GenAI Ops:
- Data preprocessing
- Model adaptation
- Serving
- Inference RAI value includes 5 categories:
- Explainability
- Bias & Fairness
- System Robustness
- Privacy & Protection
- System Security
2️⃣ Many existing solutions focus on model explainability, performance eval/monitoring, output verification, and red-teaming to address the non-deterministic nature of GenAI, and issues of bias, hallucination. More and more solutions center around security and privacy, including prompt protection, PII protection, and decentralized training.
⭐3️⃣ Lastly, there are 9 emerging RAI use cases that we’re excited about:
- Constitutional AI
- Feature Interpretability
- Optimization With User Feedback
- Data Accreditation
- Tiered Agent Permissioning
- RAG Security
- AI Compliance Explainability and Reporting
- Visual Data Understanding
- Proprietary Data Pruning
Check out the full piece here for details on the use cases we're seeing in Responsible AI: https://lnkd.in/gcbGDUPr And If you're working on solutions in this space, feel free to reach out to our team at Gradient Ventures!

