Episode summary
Daniel Guetta on the Guts of AI, Agentic AI & Why LLMs Hallucinate | The Real Eisman Playbook Ep 46
The Real Eisman Playbook
· · 1 hr 3 min · Hosts: Steve Eisman · Guests: Daniel Guetta
Summary
The episode argues that AI is best understood as two different technologies: older predictive machine learning and newer generative AI built on large language models. Daniel Guetta explains that predictive systems have been hugely valuable for years in uses like Zillow’s Zestimate, fraud detection, credit scoring, and ride pricing, but they mostly work on structured numerical data. LLMs changed the game by turning text into numerical representations, or embeddings, so models can work with unstructured language and other messy inputs.
Guetta spends much of the conversation explaining why LLMs hallucinate. In his view, they are essentially autocomplete systems that repeatedly predict the next word based on patterns learned from vast training data, so it is almost surprising when they do not hallucinate. He says this does not necessarily mean they are useless, but it does mean their output should be treated differently from human reasoning and checked in higher-level workflows.
On business value, Guetta says the biggest opportunities are not only chatbots. He groups use cases into improving traditional machine-learning systems with LLM-derived text understanding, building agentic AI that can take real-world actions like sending emails or processing returns, and using chatbots directly for writing, research, code, and document search. He is skeptical that today’s scaling path automatically leads to human-level intelligence, but he argues there is already significant practical value if companies have the data and systems in place to use AI well.
Key takeaways
- Daniel Guetta says large language models are only one part of AI; older predictive machine learning remains massively important in credit, fraud, pricing, and Zillow-style valuation models.
- Guetta argues LLMs hallucinate because they are next-word prediction systems, so getting the right answer is the surprising outcome, not the default.
- He says the biggest near-term enterprise wins come from combining LLMs with classical machine learning, especially for text-heavy tasks like content moderation and document search.
- Guetta defines agentic AI as “a large language model. A chatbot with a pair of hands,” meaning software that can actually take actions through connected tools.
- He thinks software vendors and consultants may face pressure, but he also says many companies still lack the data plumbing and workflow discipline needed to capture AI value.
Notable quotes
“large language models have almost become synonymous with AI in people's minds, but they're not”
“What you should be surprised by is that they ever don't hallucinate, right? The fact they do hallucinate is kind of, you know, that's like not the surprising part”
“These large language models are a kind of probabilistic AI”
“Agentic AI is really quite simple. It's basically just a large language model. A chatbot with a pair of hands”
“there is still so much value that you can capture”
How the episode opens
Hey, it's Steve Eisman and we're going to talk about AI today with our guest. This is a topic that we have explored for many, many months. It is crucial to the future of the US economy.
Recently, hyperscalers have announced that they are increasing their budgets enormously. The top four hyperscalers are going to literally spend $650 billion dollar alone on tech stuff all related to AI. The future of the US economy is really at stake.
And a few weeks ago we had on Gary Marcus who's a big critic of LLMs and argues that they are losing their efficaciousness. And so I wanted a second opinion because this is just too important a topic. And so today we have Columbia professor Danny Guetta who as you'll see agrees with Gary on certain points but disagrees on others and we're going to explore all of that and afterwards I'll be back to talk about what we've learned.
Hi, this is Steve Eisman and welcome to another episode of the real Eisman playbook. And today we're going to explore the world of AI, but not per se from a business perspective, but from the within the guts of AI, you know, is AI a bubble? Is it going to change the world?
You know, these are questions that we've explored over in our podcast over the last many, many months. And to help us go further is our guest, Professor Daniel Gua, who was a professor at the Columbia Business School. Daniel, welcome.
Thank you so much for having me, Steve. This is going to be great. Really excited for the conversation. really excited. So before we get started, why don't you just give us a little bit about your background to explain why you're here?
Yeah.
From the episode’s captions: roughly the first two minutes, lightly cleaned. Keep watching on YouTube →
Topics
- ai llms
- llm hallucinations
- agentic ai
- corporate adoption of ai
- ai business value
Watch the episode
Summary, takeaways and quote selection by Octagon. Quotes are short excerpts; the full conversation is in the episode.
More from The Real Eisman Playbook
-
The AI Debate is No Longer Boom vs Bust: Why It’s Time to Go Back to Stock Picking | The Weekly Wrap -
The Midterms Have Changed the AI Narrative: S&P Down 4.4%, Nasdaq Down 5.5% | The Weekly Wrap -
Why Young Americans Feel Locked Out with the Lemonade Stand Podcast | The Real Eisman Playbook Ep 82 -
Anthropic’s Growth is Slowing: Is AI Capex Next? | The Weekly Wrap -
Inflation, The Deficit & Why the Dollar Stays King with Benn Steil | The Real Eisman Playbook Ep 81 -
AI Narrative Switched, Debt Yields Spike | The Weekly Wrap
The Real Eisman Playbook, this episode and its artwork belong to Steve Eisman. Octagon is not affiliated with the show; this page is an independent summary.
