Ethan Mollick’s Post

I did a Financial Modeling and Valuation Analyst program at Corporate Finance Institute where we learned to do all this in painstaking detail with custom excel formulas that were so long as to circumfere the earth. That education is still useful and sometimes mandatory as you need it to know what questions to ask, and can sense if something simply feels off by a mile in the AI output. But obviously this will make huge swathes of analysts redundant very quickly. Or, if you look at it more positively, it will instantly promote each of them to their own managers, who can now delegate these arduous tasks to the AI and use their energy to think about the bigger picture and the business context.

While this approach might suffice for straight-forward analyses, my experience leading business ops and analytics teams has me approaching with caution. Even seemingly simple questions often reveal layers of complexity involving nuanced business logic and intricate data lineage once you look deeper. In more complex scenarios, it's not uncommon for even seasoned analysts to misinterpret data, resulting in misleading analytics that appear plausible but are fundamentally flawed. Given these challenges, I'm concerned about the potential inaccuracies an AI agent might generate, leading to even more elaborate misrepresentations.

My new favorite prompt -> “make it good” 😂

Hey newcomers, here's a tip: Begin familiarizing yourselves with the terminology and use cases for advanced modeling and analysis. In the future, it'll be more about understanding what to ask for rather than knowing how to do it yourself. Keep learning and stay curious.

I have not examined the results fully, but this seems very promising.

Ethan Mollick I've been using Claude Pro, minus the table production, since Opus came out. It's about as reliable as AI can get right now with text to data production. Even before the table add-on, Opus did a pretty good job of taking a verbal prompt and creating a spreadsheet with pivot tables that could easily generate the charts with little manual entry. The problem is it takes a lot of prompts to get something right, and even then, it would occasionally omit certain data from the manual inputs. For example, I was working on an employment law case involving back pay claims under the FLSA, and I had several payroll reports, time sheets, etc. and I was trying to get Claude to consolidate the data and create custom fields using formulas that would prove or disprove my legal theories related to the case. The bottom line is that it eventually got the job done, but it would have saved me time just to use the tools already in MS Excel and manually synthesize the data.

Ethan Mollick it would be interesting to ask your opening question: “here's excel of my startup's finances, make a dashboard." But then to cycle through the generic framework below (inductive deductive and abductive thinking) to see how different the outputs would be. Would The GPT use the context and automatically initiate the kind of structure you gave it.; 1. Tell me what you know Iterative feedback loop question 2. Tell me what you don’t know Iterative feedback loop question 3. Tell me what’s hidden Iterative feedback loop question 4. Now tell me what you think. Iterative feedback loop question 5. Now Trace the precedents thus far 6. Summarise and analyse them 7. Now help keep frame a better question.

No code AI-Powered React hosting will benefit from this trend. If regular humans can generate code like this, copy and paste it into a field and have a working app 🔥 GOODNESS KAYODE Olusegun Odufuwa

Have you tried Julius AI, I have found it be a great tool for financial anylsis.

Good point!