Eric Stylemans’ Post

Google DeepMind introduces their Frontier Safety Framework - a set of protocols for proactively identifying future AI capabilities that could cause severe harm and putting in place mechanisms to detect and mitigate them.  #AI #DeepMind #GenAI https://lnkd.in/eBSaERNB

It's great to see some of the bigger players are looking towards assessing future risks of frontier models. The idea of defining critical capability levels and using these to identify when AI systems are approaching a point where they have the capacity for substantial harm can go a long way to making sure we are responsibly mitigating the risks involved with misuse of deployed algorithms. https://lnkd.in/gSKG32wV

AI capabilities are limitless.

The culmination of lots of conversations with customers, suppliers and in the team. 🙌 Key push backs on getting on with using Gen AI are: its not safe, its too complicated and the costs are unknown. So, we've launched a fixed fee Gen AI landing zone deployment in your tenancy. It comes with everything ready for your use case and outcomes (we can also help with those!) and is safe and secure for your valuable information. Have a look: checkout our website and put your details to get a free demo, and possibly a free morning tea! 😁 #dataengine #genai #doit

I'm excited to share an important development in AI safety that I believe should become a new buzzword in our field: Critical Limit Levels (CLLs) bookmark this in your mind. This concept, introduced by DeepMind in their Frontier Safety Framework, identifies when AI capabilities reach thresholds that could pose significant risks. CLLs are a proactive measure to ensure that as we push the boundaries of AI, we do so responsibly. It's heartening to see such work being done to align AI advancements with safety and ethics. Let's spread the word about CLLs and make sure we're all on board with these crucial safety standards. What are your thoughts on integrating safety measures like CLLs in AI development? #AI #EthicsInTech #DeepMind #ArtificialIntelligence #Innovation #CLLs

Looking back, even 2019 feels like Stone Age now. The evolution of foundation models is picking up fast and we have only seen the tip of the iceberg so far. It's crucial that active conversations between industry, academia, and government happen to ensure trust and safety are not the afterthought, but key design principles of the foundation models. That's why we joined the Frontier Model Framework to help promote AI safety. https://lnkd.in/g3qgMHNw

Should you simply deploy a new AI model and leave it to do its job unattended? 🧐 Absolutely not! The world becomes more complex every day, new skills are constantly needed, meaning AI and ML models need regular monitoring, retraining, debiasing, and maintenance to ensure sustained impact. That’s what our QuantumBlack LiveOps team was made for! 💪 ⚙️ QuantumBlack’s AI-operations-managed services and products ensure long-term impact for deployed models and AI-driven solutions. Learn more about the QuantumBlack LiveOps team and QuantumBlack Coda in a Medium blog post: https://lnkd.in/dfmXjCWS #AIbyMcKinsey #QuantumBlack #ArtificialIntelligence #Digital #AIOperations

I'm proud to share the launch of our Frontier Safety Framework at Google DeepMind — a crucial step in our ongoing commitment to AI responsibility and safety. This set of exploratory protocols is designed to identify future capabilities that could pose significant risks and develop strategies to address them. By preparing today, we aim to ensure that the future of AI remains aligned with human values and societal goals. Read more in my blog post with Anca Dragan and Allan Dafoe about how we are integrating responsibility and safety at every step of the way: https://lnkd.in/evr3bjU7

🤖 AI is evolving, and so are the ways we measure its performance. From fairness to accessibility, our experts are diving deep into the social and technical aspects of #AI to address potential risks. Microsoft Research’s team combines knowledge from social sciences and technical disciplines to create tools like #AzureAI Studio safety evaluations, which continuously monitor and assess AI system responses. 🔗 Read more to explore the crucial work behind safer, fairer AI solutions: http://msft.it/6047WgCxB #MicrosoftHK #ResponsibleAI #ExplainingRAI

Are We Hyping Up LLMs? A recent article by the CEO of DeepMind caught my eye (https://lnkd.in/gaF6pvt2). It highlights the concern of inflated expectations, particularly around LLMs. LLMs are undeniably powerful tools, capable of generating human-quality text and completing complex tasks. But are we misleading the public by calling them "AI" in the sense of a thinking being? John Searle's Chinese Room thought experiment famously argues that computers can manipulate symbols without understanding their meaning. LLMs process information similarly, raising questions about their true level of "intelligence." Here's my question for the tech community: Are LLMs simply efficiency tools, revolutionising how we handle information, or are we incorrectly overstating their capabilities? Do you agree?