AI Didn‘t Start with LLMs | Shashidhar Thakur (VP/GM, Google)

Listen to full episode:

Summary: Search, answers, and actions are converging fast. And the teams building them are rewriting how information gets found, understood, and used. Google veteran Shashi Thakur joins Anant and Ed to break down what actually changed beneath the search box, why multimodal AI is a bigger leap than most people realize, and how the move from retrieval to agentic action is changing everything.

Chapters:

00:00 - Introduction and episode overview

02:02 - Ed's new role at DeepGram and AI interaction layer

04:01 - Debate on AI doom and the future of language models

07:58 - Guest introduction: Shashi Thakur and his role at Google

08:52 - The evolution of search from keyword to knowledge graph

12:06 - Understanding language learning and context in NLP

13:57 - Early search engine techniques and the rise of statistical models

18:00 - The advent of neural models and deep learning in search

22:01 - Transformers, Word2Vec, and embedding techniques

27:02 - The current state of AI evaluation and quality metrics

35:52 - Multimodal AI and the integration of images, voice, and text

43:55 - Agentic AI, recursive retrieval, and automation protocols

55:02 - The future of AI in search, actions, and enterprise applications

Sound Bites:

"Language learning is embodied in reading and context."

"Transformers and embeddings revolutionized search."

"Evaluation of AI quality is now a statistical, principle-based process."