Generative AI, explained by the people building it.
Future of Tech pairs Avishai Sharlin, Division President of Amdocs Technology, with technology leaders for conversations about generative AI. Guests such as NVIDIA CTO Michael Kagan, Prof. Yoav Shoham and Dr. Sharon Zhou talk about training data, task-specific models and who owns a company's LLM.
PRESS PLAY
Future of Tech is an interview show hosted by Avishai Sharlin, Division President of Amdocs Technology. Its introduction covers a lot of ground, from machine learning aimed at the world's biggest problems to AI that might help with biological bottlenecks in human thinking. The episodes in this sample come from a special season on generative AI, with guests who work where it is being built. Each one is a conversation between the host and a single guest. Michael Kagan, chief technology officer of NVIDIA, has said that generative AI is having its iPhone moment, and the conversation works out what he means by that. Kagan's clearest image is a loose comparison to Windows. Windows gives programmers a place to write and run C code, and in the same way a large language model becomes the interface that lets far more people program a computer. Sharlin asks how NVIDIA's Nemo compares with other language models and when Kagan realized generative AI was taking off. Nemo, Kagan explains, is a large language model tuned and optimized for NVIDIA's own architecture. Prof. Yoav Shoham of AI21 offers a way to compare models. Scale is one measure: 300 billion training tokens once seemed like a lot, and now the field works in the trillions. Data quality is another, with natural text from the internet, Wikipedia and books alongside synthetic data. Specialization is the one Shoham backs, pointing to the task-specific models AI21 has pioneered for jobs like summarizing text or answering questions in a retrieval-augmented style. By Shoham's account, those models tend to beat even the best general-purpose models at their tasks, and because they are small, they stay fast and cheap to run. Dr. Sharon Zhou comes at the subject from the enterprise side. Zhou's startup began before ChatGPT, in response to strong demand from companies already using language models that needed help. Ownership comes first: a company that only calls an external API owns nothing and has no moat, because the model itself is the new IP. Personalization comes next, with companies wanting control over fitting a model to data spread across complex silos. In these conversations Sharlin's questions stay practical: what is unique about a business, and why was it started? The talk gets technical but stays easy to follow, because the guests explain as they go. Episodes close warmly, and listeners are invited to contact the host directly on LinkedIn. Where to start depends on the question: Kagan for AI as a computing platform, Shoham for model design and cost, Zhou for how companies take ownership of LLMs.
Made for: Suits a listener who follows generative AI with an eye on strategy and wants builders at places like NVIDIA and AI21 to explain their choices, from model size to data ownership, in plain conversation.
What sets it apart: The interviewer is a technology executive, Division President of Amdocs Technology, and he keeps asking about the business case: what makes a startup distinct and when generative AI took off. Guests answer with specifics such as token counts, task-specific models and the language model as a programming interface.
How will machine learning, AI, hybrid clouds, IOT, open source, edge computing, and many other innovations solve some of the world’s biggest challenges? Sit down with some of the most innovative minds in technology to learn how they are disrupting the present, and what kind of impact they foresee for the future. Hear from companies like Google, AWS, Intel, Red Hat, HPE, Dell and new, hot startups. Hosted by Avishai Sharlin, Division President of Amdocs Technology.
Short reviews from the PlayNext desk, based on the episodes we processed.
A grounded look at how scale, data quality, specialization, and business realities shape language-model design.
Yoav Shoham lays out a clear view of the forces reshaping language models, from the jump between 300 billion tokens and multi-trillion-token training to the divide between natural and synthetic data.
A clear look at enterprise LLMs through ownership, data control, compute, and talent.
Enterprise AI is framed here as a question of ownership and infrastructure, not just a race to plug external LLMs into knowledge work. Sharon Zhu links the obvious appeal of these models to the tougher enterprise problems beneath them: building internal skills, adapting systems…
A clear, candid look at NVIDIA's GenAI turning point, from transformer momentum to NeMo's Windows-like interface.
Future of Tech uses Michael Kagan's view that GenAI is having its "iPhone moment" as a way into the systems underneath: NVIDIA's architecture, its tuned language model, and the pace at which public adoption reshaped industry expectations.
Reality, examined past the shortcut.
Global news, sharpened into consequence.
Tech power, counted in costs and constraints.
TED's stage, one speaker at a time.
Philosophy with technical consequences.
An interview with AI pioneer Yoav Shoham about the future of generative AI and large language models.
An interview with AI expert Sharon Zhu about the impact of generative AI and emerging technologies.
Avishai Sharlin interviews NVIDIA CTO Michael Kagan about generative AI's 'iPhone moment' and its challenges and opportunities.