Aki Ranin

This is what AI 'thinks' of my interview with Prof. Kevin Werbach 😅: "David Hardoon delivers a masterclass in Accountable AI! From Singapore’s FEAT principles to Standard Chartered’s AI innovation, he proves governance is a catalyst for trust and progress in fintech. A must-watch for anyone shaping AI’s future—accountability isn’t a burden, it’s a competitive edge! 🚀" I think AI is being far too kind, or maybe its just AI particularly agreeing with the core theme that one can't outsource accountability in AI to AI 😁 Key Takeaways: Justify, Don’t Blindly Trust: AI decisions must be rigorously interrogated to ensure fairness and transparency. Governance Drives Innovation: Embedding accountability in AI processes boosts resilience and value, not bureaucracy. GenAI’s Dual Edge: Harness its potential to fight scams, but only with robust, human-led oversight. Global Lessons: Singapore’s collaborative AI policies (like Veritas) set a benchmark for balancing regulation and innovation. https://lnkd.in/gUVAyrWx

116

8 Comments

The biggest barrier to widespread #GenAI and #AgenticAI #adoption isn't just model accuracy, it's economics!! Specifically, the massive cost of inference at scale. #Inference is the "live" part of AI. It’s the computational cost of generating a response from a trained model. While training is a one-time cost, inference is a continuous, operational expense that can cripple budgets. How does a company build a durable advantage here? Amazon played the long game. Their 2015 acquisition of #AnnaPurna #Labs gave them in-house expertise to build #custom #silicon. Chips like AWS Inferentia are not general-purpose, rather they are purpose-built to do one thing exceptionally well and that is to run AI inference efficiently. By controlling the hardware stack, Amazon Web Services (AWS) can optimize for performance and cost in a way that competitors who rely on third-party hardware cannot. It’s a crucial edge in the race to make AI economically viable for everyone. #AIStrategy #CloudInfrastructure

80

12 Comments

One of the great privileges of my role at STAC is getting to partner directly with leading exchanges like SGX Group, ASX, and Japan Exchange Group. These partnerships allow us to do deep research into the most pressing technical challenges in financial market infrastructure. STAC is a unique forum where fiercely competitive financial firms come together to share many of their research breakthroughs for the betterment of the industry. Hosting these events directly at the exchanges is both symbolic and practical for direct technical collaboration. Our current research areas include: - Fairness throughout the tick-to-trade cycle - Quantum particles for time synchronization - Reducing AI inference latency Our AI work is highly focused and problem-driven, and we’re quickly becoming the go-to hub for applied AI in finance, with so many experts from the world’s largest financial firms set to join us at these exchange venues to share insights from the advanced AI projects they’re working on. We are actively researching: - Context engineering best practices - Quality measurement beyond gameable benchmarks - Semantic drift in API-based LLMs The research offers an objective foundation for understanding performance, identifying areas for improvement, and navigating the technical demands of capital markets infrastructure alongside the rapidly evolving use of AI in exchange-based trading. I'm excited to head out to the ASX in Sydney and JPX in Tokyo this September, and look forward to sharing news about additional exchange partnerships very soon! Register here to be part of the action 😊 https://lnkd.in/eXPJf8Sy

54

𝗔𝗜 𝗔𝗴𝗲𝗻𝘁𝘀 𝗖𝗼𝗻𝗰𝗲𝗽𝘁𝘀 - 𝗧𝗵𝗲 𝗕𝘂𝗶𝗹𝗱𝗶𝗻𝗴 𝗕𝗹𝗼𝗰𝗸𝘀 𝗼𝗳 𝗡𝗲𝘅𝘁-𝗚𝗲𝗻 𝗜𝗻𝘁𝗲𝗹𝗹𝗶𝗴𝗲𝗻𝗰𝗲 Artificial Intelligence agents are rapidly evolving into complex systems that go far beyond simple prompts and responses. They combine reasoning, autonomy, communication, and adaptability to function in real-world environments and enterprise ecosystems. Here are the seven key pillars of AI agent concepts: 𝟭. 𝗟𝗟𝗠 𝗙𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻  The core layer where tokenization, inference parameters, embeddings, fine-tuning, and prompt engineering form the basis of agent intelligence. This is where the raw capability of large language models is shaped into usable intelligence. 𝟮. 𝗔𝗴𝗲𝗻𝘁 𝗖𝗼𝗿𝗲  The engine of reasoning and memory. It enables tool usage, self-consistency, dynamic context injection, and long-term vs short-term memory management. This is how agents move from static outputs to adaptive decision-making. 𝟯. 𝗔𝘂𝘁𝗼𝗻𝗼𝗺𝘆 & 𝗗𝗲𝗰𝗶𝘀𝗶𝗼𝗻 𝗠𝗮𝗸𝗶𝗻𝗴  Agents must act independently. With self-reflection, adaptive decision trees, task decomposition, and risk-aware decision-making, they can plan, prioritize, and correct themselves in real time. 𝟰. 𝗖𝗼𝗺𝗺𝘂𝗻𝗶𝗰𝗮𝘁𝗶𝗼𝗻 𝗟𝗮𝘆𝗲𝗿  Agents interact not just with humans, but with other agents and systems. This layer ensures natural language interaction, multimodal input-output handling, API integration, and inter-agent protocols. Effective communication makes collaboration possible. 𝟱. 𝗖𝗼𝗼𝗿𝗱𝗶𝗻𝗮𝘁𝗶𝗼𝗻 & 𝗖𝗼𝗹𝗹𝗮𝗯𝗼𝗿𝗮𝘁𝗶𝗼𝗻  In real-world applications, multiple agents must work together. Role assignment, negotiation, consensus, and distributed knowledge sharing allow agents to function as a team across domains. 𝟲. 𝗘𝗻𝘃𝗶𝗿𝗼𝗻𝗺𝗲𝗻𝘁 𝗜𝗻𝘁𝗲𝗿𝗮𝗰𝘁𝗶𝗼𝗻  True intelligence requires grounding in context. This layer manages integration with real-world tools, adaptive environment modeling, sensor data processing, and context awareness (time, location, user state). 𝟳. 𝗠𝗼𝗻𝗶𝘁𝗼𝗿𝗶𝗻𝗴 & 𝗢𝗽𝘁𝗶𝗺𝗶𝘇𝗮𝘁𝗶𝗼𝗻  Performance cannot be left unchecked. Monitoring tracks logging, error recovery, bias detection, scalability, and robustness. Continuous learning ensures agents remain reliable and efficient over time. These seven layers together form the foundation of AI agents that can reason, act, collaborate, and evolve. As enterprises adopt AI at scale, understanding these concepts will be essential for building trustworthy, resilient, and adaptive systems. Which of these layers do you believe will play the most crucial role in driving enterprise adoption of AI agents in the next few years? Follow Dr. Rishi Kumar for similar insights! --- 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 - https://lnkd.in/dFtDWPi5 𝗫 - https://x.com/contactrishi 𝗠𝗲𝗱𝗶𝘂𝗺 - https://lnkd.in/d8_f25tH 𝗪𝗲𝗯𝘀𝗶𝘁𝗲 - https://lnkd.in/ghRs8vjE #AI #AIAgents #LLM #MachineLearning #ArtificialIntelligence #EnterpriseAI #FutureOfAI

129

17 Comments

MCP (Model Context Protocol) has emerged as the standard way for LLMs to connect to external tools, servers, and data sources. The protocol defines how those connections work - a shared schema for exposing resources and tool calls. That’s why it’s being adopted across OpenAI, Anthropic, LangGraph, CrewAI, Composio, and more. But here’s the reality: MCP only gives you the spec. If you’re building an app, you still need to: ▪️ tell your copilot which MCP servers to talk to ▪️ manage how responses flow into conversations ▪️ and build UI to make tool calls usable It works, but it’s a lot of wiring. That’s the gap CopilotKit🪁 (AG-UI)’s MCP Client closes. Check here: https://lnkd.in/dF9R7PrZ Point your copilot at an MCP server, and it automatically: ✔️ discovers tools and resources at runtime ✔️ makes them callable by the LLM like native actions ✔️ streams results directly into the conversation ✔️ renders them with built-in UI components Example: instead of hand-building a dashboard for SQL queries, your copilot can connect to a database MCP server, let the LLM generate queries, and then render the results in-app with a ready-made table component. No extra plumbing required. 👉 MCP defined the protocol. 👉 CopilotKit MCP Client makes it practical in your app. Setup takes one command: npx copilotkit@latest init -m MCP

419

85 Comments