Tirrell Payton

Massachusetts Institute of Technology just unveiled a genuinely impressive tool for #AIgovernance: https://airisk.mit.edu. - The Risk Database identifies 700 types of risks captured from over 43 frameworks (with full attribution). - The Casual Taxonomy of AI Risks classifies how, when, and why these risks occur (for example if the risk is intentional or not intentional). - The Domain Taxonomy of AI Risks places these risks into 7 domains (such as misinformation). I can easily see this as a common frame of reference for researchers, developers, auditors, policymakers, regulators and others. Can't wait to dig more into this and put it to good use.

4

From where I sit in the SF Bay Area, cradle of innovation for at least the last 30 years, I'm a true believer in the promise of the #AI revolution. That said, I find it fascinating to consider views on the opposing side. I'm sharing below two opinion pieces from the last two weeks suggesting that the AI revolution is fundamentally overhyped and a bubble ripe for bursting, and that AI tools are actually not useful or transformative. Contrast this with venture capitalist Kai-Fu Lee's prognostication that 50% of white collar jobs will be displaced by #AI within three years, and that OpenAI and the other cutting edge movers in the space are headed to a trillion dollar valuation. Which side will win out? Where will we be by 2030 in the AI race? I'd love to hear your thoughts in the comments.

7

3 Comments

After the launch of the PCI Data Cooperatives Report (https://lnkd.in/g_CHSvdd) on September 16, the journey continues. Yesterday I spoke on Data Cooperatives and the Digital Commons at the Harvard Club of NYC, as part of the Leaders of Regeneration series. 🌍 As we face climate change and biodiversity loss, and ecosystem collapse, threatening 40% of lives and destabilizing economies, data cooperatives offer a powerful solution. By fostering collaboration and responsible data sharing, they can drive the innovation and collective action needed to protect ecosystems, enhance economies, and create a more sustainable future. #DataCooperatives #DigitalCommons #DigitalGovernance

32

How does an enterprise boost their confidence in Generative AI? 🤔💡 As enterprises move from experimentation to production with genAI, a crucial question looms: What if something goes wrong? The fear of AI hallucinations and reliability issues is real, but innovative solutions are emerging to address these concerns. 🚀 Patronus AI, a San Francisco-based startup is making waves in the AI evaluation space. They've just closed a $17 million Series A funding round, led by Notable Capital and joined by Datadog Inc., Lightspeed Venture Partners, and Factorial Capital. 💰 But why is this significant for you and your business? Their platform promises to: 1️⃣ Detect and mitigate hallucinations 2️⃣ Test LLM reliability with adversarial prompts 3️⃣ Provide comprehensive evaluation tools for model performance Patronus AI recently unveiled Lynx, an open-source model for detecting and mitigating hallucinations in LLMs. It addresses one of the most significant challenges in AI adoption -> ensuring the reliability and safety of LLMs. Few highlights: ▪ Outperforms industry giants: Lynx surpasses GPT-4o, Claude-3-Sonnet, and other closed and open-source models in hallucination detection tasks. ▪ Impressive accuracy: Lynx (70B) achieved 87.4% overall accuracy on HaluBench, outperforming GPT-4 by 8.3% in detecting medical inaccuracies! ▪ Two Powerful Versions: Lynx comes in 70B and 8B parameter versions, catering to different computational needs while maintaining high performance. ▪ Innovative Approach: Lynx employs a novel method of generating hard-to-detect hallucination examples using semantic perturbations. ▪ Introducing HaluBench: A comprehensive hallucination evaluation benchmark with 15k samples across real-world domains, including finance and medicine. ▪ Open-source availability: Both Lynx and HaluBench are freely accessible on HuggingFace. As enterprises increasingly rely on LLMs and RAG systems, tools like Lynx could be crucial in building confidence and accelerating AI integration into critical business processes. #AI #EnterpriseAI #OpenSource #TechInnovation Read the full story here: https://lnkd.in/dAg_dN5c

6

For those implementing or thinking about implementing AI - this will be an invaluable and freely available database of risks and areas to watch out for with AI ... and from a reputable source! https://lnkd.in/e6PmTDp3

8

The Stanford Human-Centered AI Report is out. This is the 7th year that Stanford has published this report and it is a must-read if you're trying to keep track of the developments in this field. https://lnkd.in/gwsHtzsq

2

Model Explainability means different outcomes for different users. We can't have one size fits all when explaining model prediction. At the same time, unless the explainability is true to the model, it has credibility issues! All regulators expect to have explainable ML models, which is not standardised! How do you scale ML explainability in high-risk and high-compliance use cases? Continuing the article in Forbes, I've written about various XAI outcomes and how to approach setting up such explainability. #AryaXAI #ExplainableAI #EnterpriseAI #AIRegulations https://lnkd.in/gp7YgzB3

13

1 Comment