The AI-in-Finance Landscape
Finance was one of the first industries to adopt machine learning at scale — algorithmic trading desks were using statistical models decades before ChatGPT existed. But the arrival of large language models changed something fundamental: for the first time, the unstructured language of finance — earnings calls, filings, analyst notes, news — became something software could read and reason about. This module maps the real landscape so you understand what AI does well in finance, what it does badly, and where the genuine value is.
Two very different kinds of "AI in finance"
When people say "AI in finance," they're usually conflating two completely different things. Understanding the distinction is the foundation of everything else in this course.
The first is quantitative machine learning — the statistical models that have powered algorithmic trading, fraud detection, and credit scoring for years. These are specialized, narrow systems trained on structured numerical data. A fraud model doesn't "understand" anything; it detects statistical anomalies in transaction patterns. This world predates LLMs entirely and largely operates separately from them.
The second — and the focus of this course — is the use of large language models like Claude, ChatGPT, Gemini, and Perplexity to work with the language of finance. Reading a 10-K. Summarizing an earnings call. Drafting an investment memo. Researching a company. Comparing competitors. This is new, and it's where most professionals are now spending their AI energy.
This course is about using general-purpose LLMs as a research and analysis assistant — not about building quant trading algorithms. The skills here apply whether you're an analyst, an investor, a finance student, or someone managing their own portfolio. You don't need to code. You need to understand how to direct these tools and, critically, how to verify what they produce.
Where LLMs genuinely add value in finance
Be skeptical of anyone claiming AI will replace financial analysts. The reality is more specific and more useful: LLMs are extraordinarily good at a particular class of tasks that used to consume enormous amounts of an analyst's time.
Where LLMs are dangerous in finance
Finance is a domain where being confidently wrong has real consequences — money, fiduciary duty, regulatory exposure. The same model that summarizes a 10-K beautifully will also invent a revenue figure that sounds completely plausible. You must internalize where the failure modes are.
Fabricated numbers. LLMs hallucinate specific figures — revenue, margins, ratios, dates — with total confidence. A made-up "14.2% operating margin" reads exactly like a real one. Never trust a number an LLM gives you without checking the source.
Stale knowledge. Models have training cutoffs. Ask about a company's "current" CEO, latest quarter, or recent stock price from training data, and you may get information that's months or years out of date — stated as if it's current.
False precision in reasoning. An LLM can produce a confident-sounding valuation or recommendation built on flawed assumptions it never flagged. The polish of the output masks the weakness of the foundation.
The tool-to-task map for finance
You learned the individual tools in the earlier courses. Here's how they map specifically to financial work.
| Task | Best tool | Why |
|---|---|---|
| Current market data, recent news, latest filings | Perplexity | Live web search with citations defeats the training-cutoff problem |
| Analyzing a long filing or transcript you upload | Claude or Gemini | Large context windows handle full documents; Claude for nuance, Gemini for sheer length |
| Running actual calculations on a dataset | ChatGPT (Advanced Data Analysis) | Executes real Python on uploaded spreadsheets — real numbers, not estimates |
| Writing the investment memo or thesis | Claude | Highest-quality analytical writing and willingness to flag weak arguments |
| Comprehensive multi-source company research | Gemini Deep Research or Perplexity | Agentic multi-search synthesis across many sources |
Treat AI as a brilliant, fast, occasionally unreliable junior analyst. It can do the legwork — read the filings, draft the memo, build the framework — but everything it produces requires senior review before it's trusted. You are the senior. The entire value of this course is teaching you to direct the work and catch the errors.
What's ahead
The next module tackles the most important technical reality of all: how language models actually handle numbers, and why that determines everything about how you should and shouldn't use them for financial math. From there we move through research, statement analysis, modeling, filings, valuation, risk, building your own workflow, and finally the ethical and practical limits that every responsible practitioner must respect.