Why IBM Is Betting Everything on Small AI Models

In this episode of Eye on AI, Craig Smith sits down with Sriram Raghavan, Vice President of AI at IBM Research, to explore one of the most important debates in enterprise AI right now. Do you actually need a massive model to get world class results? IBM's answer is no, and Sriram breaks down exactly why.

Sriram explains why IBM chose to train its Granite models directly using reinforcement learning rather than distilling from larger models like most of the industry. The reason goes beyond performance. It comes down to data lineage, safety alignment, and a belief that small, efficient models are the only sustainable path for enterprises running AI across hybrid cloud environments.

We get into the full technical stack behind that bet. How data quality has replaced model size as the real competitive advantage. Why parameter count is becoming the wrong metric entirely. How IBM's inference time scaling techniques allow an 8 billion parameter model to match the performance of GPT-4o and Claude 3.5 on code and math benchmarks. And why IBM is pioneering a new concept called Generative Computing, which treats AI models not as prompt receivers but as programmable computing elements with runtimes, modular LoRA adapters, and proper programming abstractions.

Sriram also shares where IBM Research is headed next, including breakthroughs in continuous learning, agent orchestration, and making unstructured enterprise data actually usable at scale.

Subscribe for more conversations with the people building the future of AI and emerging technology.

Stay Updated:

Craig Smith on X: https://x.com/craigss

Eye on A.I. on X: https://x.com/EyeOn_AI

(00:00) Why IBM Skips Distillation and Trains Small Models Directly

(04:50) Did We Even Need Giant AI Models in the First Place?

(08:12) How Data Quality Became the New Competitive Moat

(11:54) Why Parameter Count Is the Wrong Way to Measure a Model

(15:36) Reinforcement Learning Without Losing Broad Capabilities

(22:05) Inference Time Scaling: Getting Big Model Results From Small Models

(28:12) Generative Computing: Treating AI as a Programming Element

(36:40) Why IBM Open Sources and How Small Models Make It Sustainable

(41:25) The Path to Continuous Learning Without Rewriting Weights

(51:00) IBM's Full Roadmap: Models, Data, and Agents

What if the country that trained the world's engineers finally decided to keep them?

In this episode of Eye on AI, Craig Smith sits down with Abhishek, the civil servant leading India's $1.2 billion national AI Mission, to explore how one of the world's largest and most diverse nations is mounting a serious challenge to US and Chinese dominance in artificial intelligence.

Abhishek breaks down the honest story behind India's late start. World-class talent, but no research ecosystem to retain it. Digitization without AI-usable data. Compute so scarce that the entire country had fewer than 500 GPUs just two years ago. And a brain drain so severe that the engineers India trained are now running the biggest tech companies in the world, just not from India.

Abhishek walks through exactly how the mission is tackling each of those gaps. A subsidized compute program that gives researchers and startups access to 38,000 GPUs at under a dollar per hour. AI Kosh, a national data platform pulling public and private sector datasets into a single AI-ready repository. Centers of Excellence connecting IITs around domain-specific research in agriculture, healthcare, education and mobility. And a sovereign LLM program, with four models already in development and eight more on the way, built specifically for India's languages, voices and needs.

We also get into the geopolitics. Where India stands as the US and China carve out competing AI spheres of influence. Why Abhishek is pushing for a UN-led governance framework rather than aligning with either bloc. And what it would actually take for a country of 1.4 billion people to not just catch up, but leapfrog.

Subscribe for more conversations with the people shaping the future of AI and emerging technology.

Stay Updated:

Craig Smith on X: https://x.com/craigss

Eye on A.I. on X: https://x.com/EyeOn_AI

(00:00) Introduction and Abhishek's Background

(02:43) What the India AI Mission Is and How It Started

(04:53) The $1.2 Billion Budget and Total AI Investment in India

(06:36) Data Center Build-Out and the Road to 7 Gigawatts

(08:11) AI Kosh: India's National Data Platform

(10:50) Subsidized GPUs and How Researchers Access Compute

(12:41) Brain Drain, Reverse Migration and Retaining Top Talent

(17:24) Centers of Excellence Across IITs and Key Sectors

(19:21) Expanding Fellowships and Training the Next Generation

(20:11) Why India Started Late and What Changed

(21:48) Sovereign LLMs Built for Indian Languages and Needs

(22:42) The Diversity Challenge and Culturally Relevant AI

(23:16) Government Funding for Foundation Model Development

(24:12) The AI Impact Summit and India's Role on the Global Stage

(24:52) India, China, the US and the Battle for AI Governance

(29:37) The UN Framework and India's Third Way

(31:16) India-China Relations and New AI Partnerships

(32:01) How the $1.2 Billion Budget Was Decided

(33:31) Can India Actually Catch Up With the US and China

Most enterprises are excited about agentic AI. But very few are actually deploying it in production.

In this episode of Eye on AI, Craig Smith sits down with Adi Kuruganti, Chief AI and Development Officer at Automation Anywhere, to break down why agentic AI is so hard to get right in the enterprise and what it actually takes to move from a promising pilot to a mission-critical deployment.

Adi explains why the future of enterprise automation is not agentic AI alone, but the combination of deterministic and agentic systems working together, and why companies that treat AI as a technology problem instead of a business outcomes problem are setting themselves up to fail.

They dig into how Automation Anywhere is orchestrating agents across legacy systems, healthcare platforms, and financial services workflows, why governance and compliance are the first questions every enterprise asks, and how their Process Reasoning Engine is continuously improving agent performance using metadata from over 400 million running processes.

The conversation also covers the real timeline to a fully autonomous enterprise, why the POC to production gap is the biggest failure point in enterprise AI today, and what companies that wait too long risk losing to competitors who started the journey earlier.

If you want to understand where enterprise AI actually stands today and what it takes to deploy it responsibly at scale, this episode gives you a clear and grounded perspective.

Subscribe for more conversations with the people building the future of AI and emerging technology.

Stay Updated:

Craig Smith on X: https://x.com/craigss

Eye on A.I. on X: https://x.com/EyeOn_AI

(00:00) Why Enterprises Are Struggling With Agentic AI

(02:39) What Automation Anywhere Does and the APA Category Explained

(08:01) Deterministic vs Agentic AI: Why You Need Both

(10:59) How Human in the Loop Works in Enterprise AI

(17:16) The Mozart Orchestrator and Process Reasoning Engine

(23:50) How AI Is Upgrading and Replacing Classic RPA

(27:31) How Automation Anywhere Works With Enterprise Customers

(31:53) The Biggest Challenges of Scaling Agentic AI

(41:10) The OpenAI Partnership and What It Means

(47:06) Training Staff and Building AI Literacy at Scale

(51:39) Staying Close to Customers as the Technology Shifts

(53:17) Is the Autonomous Enterprise Actually Coming