Pradeep Sanyal

Reflecting on my time at Davos last week: It’s always instructive to talk to people working at the cutting edge of technology. Open Source AI is the key to building a diverse set of AI systems and tools - it is how we will get to a point where AI tools can understand, reflect and benefit people from all languages and cultures in the world. You can watch the full ‘Debating Technology' discussion here where we delve into some of the opportunities and challenges facing us all today - https://lnkd.in/e93rKtEY

976

63 Comments

The most profound implication is what Stephen Wolfram calls 'computational inevitability.' The transformer didn't accidentally discover these conservation laws—any system optimizing for semantic coherence in high-dimensional information spaces will necessarily converge toward these same thermodynamic configurations. It was 'computationally inevitable.' This suggests we're not just building AI systems—we're mapping the landscape of computational possibility itself. The conservation constants are coordinates in this landscape, marking the locations where meaning-preserving computation becomes possible. The real narrative is that intelligence, meaning, and semantic processing are universal computational phenomena. Whether implemented in silicon, biological neural networks, or potentially even the computational substrate of physical reality itself, these same conservation principles will emerge. We're discovering that the universe has preferences for how information should be organized to generate meaning—and transformers have stumbled upon these universal computational preferences through pure optimization.

110

31 Comments

🚀 Just wrapped up a hands-on project applying Retrieval-Augmented Generation (RAG) to a real-world use case: interpreting LIC’s New Jeevan Shanti policy document using LLMs. The core idea? Combine vector search over regulatory PDFs with GPT-4 to generate grounded, faithful answers to complex policy questions. Key capabilities we built: ✅ LCEL-based RAG architecture using LangChain ✅ Contextual compression retriever for focused document lookup ✅ Modular LLM & embedding selection (OpenAI, HuggingFace, Mistral) ✅ Evaluation using LLM-powered metrics: faithfulness, relevance, correctness 📄 The document used is publicly available here: LIC New Jeevan Shanti Policy PDF 💻 Full code available on GitHub: https://lnkd.in/gS9mEzWA 💡 This architecture brings the best of both worlds — retrieval accuracy and generative flexibility — especially in regulated domains like insurance and financial services. 🔜 Next step: Incorporate Agents that can reason across multiple steps — not just answer a question, but decide which tools to use, when to search, how to validate, and when to stop. That’s where the future of enterprise GenAI lies. Would love to hear how others are pushing RAG, agents, and LLMs in production! #GenAI #LangChain #LLM #RAG #AIForOps #InsuranceTech #LLMEvaluation #LangChainAgents

60

1 Comment

Announcing AI Aspire, our new advisory firm to help enterprises with their AI strategy and transformation journey! We are partnering with Bain & Company and looking forward to helping businesses unlock scalable, transformative value. C-suite is now realizing that top-down leadership is needed for AI transformation. But figuring out the implications of AI on a particular business is extremely complex. The Bain team is world class at helping businesses craft strategy and navigate complex landscapes. Kirsty Tan (at AI Aspire) and I are thrilled to be working with Bain's Chuck Whitten, Sarah Elk, Erika Serow and team to help businesses. Questions that C-suite and boards are now asking include:  - What new products are now possible? What can we (or our competitors) now do for customers that were not possible before?  - How do we use AI to boost productivity? What workflows can be streamlined or processes reinvented?  - What technology investments, for example in data infrastructure, should now be prioritized? And what are the risks, for example to security and compliance?  - What are the HR implications: What roles do we hire more or fewer of?  - How do we bring the team with us for the journey, and enable transformation from within?  - Does AI enable new competitors; or alternatively, are there now new markets that are possible for us to move into? Technology is no longer merely a support system for business — it is  engine of growth for the companies nimble enough to adapt and daring enough to lead. Press release: https://lnkd.in/gBq7Jm4B

4,041

209 Comments

Reflections from my Notes on Credo AI Agents of Trust Summit 2025 highlights how trust has evolved from an abstract concept to a fundamental protocol driving AI value. Navrina Singh noted in her keynote, we've entered the "Age of Agency" where organizations implementing multiple AI agents simultaneously face new coordination challenges that require robust governance frameworks. ⚡ McKinsey's projection of $4.4 trillion annualized AI value underscored a recurring theme: value follows confidence, and (#AI) governance at scale amplifies this value. ⚡ Dean Ball, author of the White House #AIActionPlan discussions emphasized that "self-governance means enterprises need to clearly define what #AITrust means to them," requiring governance teams to be completely AI literate and capable of strategic AI decision-making throughout development and deployment. ⚡ The #AITrustSummit revealed how AI governance has matured from basic auditing to sophisticated evaluation and monitoring systems. Discussions from Microsoft, IBM, and Google showcased their approaches to agent identity, traceability, and observability—highlighting parallels with existing #cybersecurity protocols while addressing unique AI challenges like agent sprawl and shadow AI. Cisco's "Prompts by roles" initiative demonstrated practical approaches to democratizing AI literacy while maintaining governance guardrails. ⚡ "AI safety is a continuous journey, not a destination," reflected Dave Orr, Head of Safeguards at Anthropic, while sharing sobering insights from their August Threat Intelligence Report—reinforcing why Anthropic has prioritized safety and research since its inception. ⚡Key unsolved challenges: - Agent sprawl (Agents calling other Agents) - Shadow AI  - Traceability & Proper Evaluation in complex environment. - Building (extremely) adaptable algorithms/systems  Thank you, Credo AI and Navrina Singh, for facilitating these crucial conversations that reinforces how trust serves as both competitive advantage and an essential foundation for responsible AI advancement. We have our work cutout! Credo AI #AITrustSummit #ResponsibleAI #AIGovernance #AISafety #Innovation #InsureTech

59

Just when people were thinking that LLMs have reached a plateu, a new exciting research trend started with many industrial application, Reasoning. This is a great detailed yet accessible write-up describing SOTA Reasoning including techniques like Chain of Thoughts (CoT), Supervised Fine Tuning (SFT), Reinforcement Learning with Verifiable Rewards (RLVR), Distillation for Gemini, O*, DeepSeeks and more . Great work Cameron R. Wolfe, Ph.D. https://lnkd.in/dXpGTvhH

160

10 Comments

The AI discipline is evolving so fast that today's cutting-edge AI implementations will feel obsolete in months. OpenAI is now focusing on developing superintelligence, a significant leap beyond its current artificial general intelligence (AGI). This could potentially revolutionize problem-solving capabilities, accelerate scientific discovery, and contribute on a global scale. https://lnkd.in/g_BZ4dvZ

69

4 Comments

Why Adaptation—Not Just Scale—is the Future of AI Models What if adapting a large language model to a new task was as easy as describing the task in plain English? That’s exactly what Text-to-LoRA (T2L) enables. 🔍 The Problem: Today, adapting foundation models for specific tasks typically requires fine-tuning with carefully curated datasets and lots of compute. This is slow, costly, and hard to scale. 💡 The Breakthrough: T2L is a hypernetwork that instantly generates task-specific LoRA adapters—tiny plug-ins for LLMs—based solely on a natural language description of the task. No data. No fine-tuning. Just a sentence or two of instructions. It matters because : • 🧠 Zero-Shot Adaptation: T2L generates effective adapters for unseen tasks, outperforming many multi-task or average LoRA baselines. • ⚡ Single Forward Pass: LoRAs are produced in real-time—fast enough for real-world use cases. • 🧩 Steerability: Want your model to reason like a physicist or a biologist? Just tweak the task description. • 💻 Efficiency: Hundreds of LoRAs are compressed into one tiny hypernetwork—slashing memory and compute costs. 🚀 This isn’t just academic. It means faster deployment of AI in healthcare, finance, education—anywhere specialized reasoning is needed. 👉 Could this be the key to democratizing LLM customization? #AI #LLM #NLP #Hypernetworks #LoRA #TextToLoRA #FoundationModels #MachineLearning #FutureOfWork

54

5 Comments

“Brain-Inspired AI” Many of the biggest breakthroughs in AI come from understanding how the human brain works. Just like we don’t remember everything at once—but recall what’s relevant when needed—AI is learning to do the same. “Memory” is a vital part of AI agents. It lets them remember past interactions and use that knowledge to make smarter decisions. Memory isn’t just a feature—it’s the foundation for truly autonomous, goal-driven AI. The latest advance is “structured retrieval augmentation”—a method that helps AI quickly find and use only the most important information when needed, just like our brains do. The result? AI that’s faster, more efficient, and more human-like in its thinking. In the race toward more powerful AI, memory is a game-changer. Without memory, there’s no continuity. Without continuity, there’s no true agency. Nature has many smart systems—and now AI is catching up. Looking forward to more breakthroughs in the agentic memory space. Read more about Microsoft’s work: https://lnkd.in/g8qgkxxn #AI #AgenticAI #MemoryInAI #FutureOfAI #TechInnovation #StructuredRetrieval #MicrosoftAI

55

2 Comments

The MIT Media Lab conducted a notable study titled “Your Brain on ChatGPT,” which examined the cognitive effects of using #AI tools like GPT on brain activity during writing tasks. Over multiple sessions, ChatGPT users became increasingly reliant on the tool, demonstrating weaker memory recall of their own work and lower cognitive involvement. When ChatGPT users were forced to write without the tool later, their brain connectivity stayed lower, implying possible dependency. Just as forgetting phone numbers due to mobiles did not equate to “brain damage” or diminished intellectual capacity, similarly, adopting AI for assistance should be seen as a shift in cognitive strategy rather than #cognitive decline. The brain adapts to new tools, redistributing effort to tasks that require deeper thought or creativity while automating simpler activities through #technology

21

2 Comments

🧩 What is the Agentic AI Mesh — and why does it matter for enterprise AI? In a recent article, QuantumBlack, AI by McKinsey introduced the Agentic AI Mesh. The Agentic AI Mesh reflects a broader shift: from isolated LLM use cases toward orchestrated agent ecosystems embedded in enterprise processes. It is a modular, vendor-agnostic infrastructure layer that enables enterprises to: • Design, deploy, and scale multi-agent systems • Enforce enterprise-grade governance (authentication, access control, audit logs) • Integrate with internal systems and APIs in a secure, observable, and traceable manner • Support continuous feedback and improvement across agents in production • Maintain vendor neutrality through open standards such as A2A (Agent-to-Agent) and MCP (Model Chain Protocol) Core layers: 1. Agent Composition Layer – defines agents, roles, goals, and interaction patterns 2. Mesh Services Layer – includes orchestration, policy enforcement, logging, testing, and evaluation 3. Mesh Gateway – handles routing, security, and interoperability between external models/tools and internal systems A global bank was able to launch 100+ agents with fewer than 10 engineers in under 3 months, powered by the Mesh. It scales to hundreds of agents with minimal engineering overhead.

25

This Forbes article (https://lnkd.in/eBxYcBK3) from last week about the hallucination rates of the lastest LLM based reasoning models raises a few questions: Are we going in the right direction if the new models show hallucination rates of up to 79% in simple tests? Should we rather think about combining a set of LLMs and rules based solutions to build in check-points that identify and flag errors? Then there is this article: https://lnkd.in/eyWh9NDA A few months old already, and even at that time 60% of physicians reported that they use LLMs for decision making. To support diagnosis, identify drug interactions as well as treatment planning. Almost all of the physicians reported that they use other tools to check the output of the LLMs. Lets think about this: If the new LLMs have hallucination rates of up to 79% and physicians check their output for correctness, will they soon stop using these models? Since there would be no time saved if a lot of the results are just made up by an LLM? In my opinion, we need to think much more in specific use cases and build more than one LLM that is supposed to answer all questions in the world. LLMs are probabilistic models, while in healthcare we have a lot of fixed rules and foundational biological principles. Bringing these two things together is crucial. There are solutions emerging from startups and companies that have domain expertise and focus on a specific problem to be solved. These approaches will bring the real value to the healthcare industry! If you have a specific use case or example that you want to discuss please comment on this post or reach out.

21

1 Comment