.svg)
AI models don’t simply fail or succeed. They drift, stall and evolve. The real challenge is understanding why they behave the way they do and how to steer them with confidence.
In our latest episode of Zero-Touch Live, Rakuten Cloud president Partha Seetala joined Rakuten Symphony CMO Geoff Hollingworth to share the most important lessons from season two of Partha’s popular AI training series, A Comprehensive and Intuitive Introduction to Deep Learning. They dove straight into why the concepts and approaches covered in this course are so critical and how it can help engineers gain hands-on control over model behavior.
🎥 The replay is available now below.
As Partha explains, most engineers didn’t come up with AI. They've been focused on building in domains like networking, storage and systems software. And because modern AI techniques like transformers and large language models are still relatively new, those systems weren’t built with AI in mind. That makes foundational AI knowledge essential, not just to understand tools like LLMs, but to apply their underlying mechanisms to real infrastructure problems.
But the way AI is typically taught, Partha notes, is part of the problem:
“Most of the tutorials out there are inaccessible. Either they are very heavy in math or they just show you the concepts. They really do not tell you how those concepts connect to real-world problems.”
Season two of Partha's training addresses that gap by walking through the core mechanisms behind learning, starting with tokenization and embeddings, then moving through RNNs, LSTMs, Seq2Seq models and attention.
The conversation also touched on what makes today’s AI moment different from past hype cycles. For Partha, the key distinction is economic value: “Yes, there is hype but is it delivering economic value today? The answer is yes.” AI is already reducing the unit cost of work across industries, from marketing to technical writing to system optimization.
Yet Partha draws a clear line between scientific breakthroughs and engineering progress. Most of the current gains in LLMs, he says, come from engineering work like optimizing architectures, managing compute constraints and combining known techniques in more efficient ways. And that means engineers with foundational fluency have an opportunity to drive innovation forward, not just consume it. Eventually, they begin to develop the understanding to build smarter systems in their own domain.
