Demystifying AI | Using Generative AI in Data Analysis - Part 2

Join Lucas and Claire in this engaging discussion about the transformative power of generative AI in data analysis. They explore practical applications, system architectures, and the ethical considerations of AI, with insights on how AI can revolutionize various fields including self-driving cars and medical research.

Transcript:

00:00 — New Enterprise Features in Generative AI

Claire:
So with ChatGPT, they have this new Enterprise product where a company can come in and use it internally. You can upload documents like your 10-K and all of that. But is this something you have to keep updating manually?
Or can the large language model eventually figure out where to pull this data from and give it to you? Will people always have to manually enter all of this documentation?

Lucas:
It depends on the system architecture and how you set it up. For example, I set up a system on my own drive with my documents, and it automatically indexes them and applies embeddings.
But yes, I still have to put documents in there — it doesn't just pull things out of my email automatically.

You can set it up that way, though. You could say:

"Automatically add any attachments from my email,"
or "Automatically ingest email content or chat logs,"

It really depends on what access you want your enterprise AI system to have.

You also have to think about architecture choices where the LLM is only one component of the solution. Generative AI (like GPT-4 or Gemini) may interpret what the person is asking, but another system — for example, a SQL engine — may actually run the query.

You generally want guardrails so the generative model isn’t hallucinating or “imagining” things in places where precision matters.

02:26 — Decision Making and Guardrails in AI Systems

Claire:
Earlier you mentioned patterns the model can figure out, and how you can give it a broad scope or restrict it. I’m curious about your thoughts on AI making decisions, specifically in things like self-driving cars.
How does that actually work?

Lucas:
A lot of the questions around that go straight into the ethics of AI — who is responsible, and how do we compare human vs. machine decision-making?

Most real systems today are “human-in-the-loop.”
Humans constantly make mistakes — sending the wrong email, saying something the wrong way, or in the worst case, causing car accidents. A computer making a decision also might make the wrong one.

You can design a system to take a huge number of variables into account, but eventually a decision must be made — brake, turn right, turn left. That final action has consequences.

That’s why regulators still require a human with hands on the wheel.
And it raises the question: Who is responsible?

You can’t build a computer that’s perfect and always right. Humans aren’t always right, societies change their minds, ethics shift, and what’s considered “right” is fluid.

Even with 500 billion variables, we won’t build something infallible — and humans aren’t infallible either.

05:44 — Military, Safety, and AI Applications

Claire:
That makes sense, and it’s interesting because I see a lot of discussion about the military and how AI is going to improve safety. The military is already so advanced…

Lucas:
Yeah — and here’s a personal tangent:
As humanity, we spend enormous resources on things like military competition and social media marketing, essentially trying to influence each other.

But there are areas where AI could make far more meaningful contributions:

Medical research
Physics
Scientific discovery

Take MRI research, for example. Life expectancy has gone from ~75 to ~85 over a century — progress, but still limited given the massive effort invested.
Or physics: the last major breakthrough in our understanding of gravity was Einstein — a long time ago.

There are enormous scientific frontiers where humans could apply themselves more deeply. AI may help unlock breakthroughs by exploring molecules, simulating physical systems, and accelerating discovery.

07:09 — AI in Medical Research and Physics

Lucas:
There are incredibly interesting scenarios in medicine, physics, and molecular science where AI can be applied — and I think that’s where it becomes truly exciting. AI can help with discovery, experimentation, and testing in ways that could meaningfully advance humanity.