Introduction

Show Notes:

In this episode, Will Bachman talks to Adam Braff, a former McKinsey partner who specializes in data analytics. Adam has been using chat GPT to explore how this powerful tool can be harnessed for data analysis. He explores the implications and potential impact of this innovative approach.

The Quest for Analyzing Quantitative Data

The ability to analyze quantitative data using generative AI has long been a holy grail for many data scientists. While Chat GPT and other language models have proven their prowess in generating text and even creating visual content. Adam talks about how to  tackle the challenge of applying these tools to analyze large datasets problems and uncover potential solutions.

Adam outlines four key aspects of the problem at hand. First, there is a need to upload data into the Chat GPT tool, as the existing training data may not encompass the specific dataset of interest. Second, an intuitive interface is required to facilitate a conversation with the tool, allowing for iterative exploration and analysis. Third, the ability to visualize the data in various formats, such as tables and graphs, is crucial for understanding and validating the results. Lastly, incorporating up-to-date contextual information about the world around us is essential to gain insights into correlations and patterns within the data.

Uploading Data: Bridging the Gap

To address the challenge of uploading data into Chat GPT, several options have emerged. One approach involves integration with popular spreadsheet tools like Google Sheets and Microsoft Excel. Users can interact with the data by writing formulas and commands directly within the spreadsheet software. 

Another option is to paste data directly into Chat GPT, as long as it fits within the context window. This approach allows for a quick overview of the data and initial exploration of its contents. The ability to have a conversation with chat GPT is a significant breakthrough in data analytics. Adam highlights the emergence of third-party plugins that enable users to interact with the tool directly. These plugins, such as "chat with your data" and "chat with G sheet," bring us closer to the goal of conversational data analysis within the chat GPT environment.

Additionally, separate startups have leveraged APIs to connect with open AI models like GPT 3.5 and GPT 4. These startups, such as seek.ai and data DM, provide an alternative approach to interact with the data, although they operate outside the chat GPT window.

Code Interpreter: The 800-Pound Gorilla

Among the various solutions, Chat GPT code interpreter stands out as a powerful tool for data analysis. As an official open AI product, it offers a native and robust interface within Chat GPT. By activating code interpreter, users gain access to a chatbot-like interface where they can upload data, ask questions, and receive answers in real-time.

The code interpreter translates user queries into Python code, allowing for complex data manipulations and analyses. For example, if a user wants to analyze the correlation between variables or observe trends over time, code interpreter can aggregate and analyze the data accordingly. While the current interface may require users to refer back to the original spreadsheet for column names and other details, it provides a promising solution for non-technical analysts to engage with data.

Unleashing the Potential: A Case Study

To illustrate the capabilities of code interpreter, Adam conducted an analysis using three datasets: daily credit card spending on fast food brands, weekly food spending in various categories, and macroeconomic data from the Federal Reserve. The goal was to explore correlations between fast food spending, overall food spending, and economic conditions.

By uploading these datasets into code interpreter, Adam engaged in a conversation with the tool, asking questions and receiving insights on trends overtime. The analysis aimed to uncover potential drivers of spending on fast food brands and identify correlations with broader food spending and economic indicators. Adam explains the various types of analysis and data the tool can deliver and how it can be delivered.

Accessing a Python Interpreter

For those unfamiliar with Python programming, Braff provided guidance on how to access a Python interpreter. He suggested using platforms like Replit, which allow users to create a free environment for running Python code. Additionally, he mentioned that AI language models like ChatGPT can generate Python code for specific tasks, making it easier for non-technical users to experiment with programming. He emphasizes the importance of hands-on experimentation and encourages individuals to explore these tools to enhance their data analysis skills.

Navigating the Landscape of AI Tools

Adam talks about the landscape of AI tools and their potential applications in organizations. He talks about how he experimented with scraping. He stresses the need for a problem-solving framework and highlights the importance of breaking down complex problems into manageable steps. By understanding which parts of the problem-solving process AI tools excel at, users can leverage these tools effectively. Braff also emphasized the importance of experimenting with different modalities of interaction, such as step-by-step queries or end-to-end analysis, to find the most suitable approach for each problem.

Implications and Future Impact

The ability to analyze data using chat GPT and similar tools has significant implications for various industries. Adam talks about the problem of hallucination, where the tool is limited, and how far it is to becoming a plug and play data scientist. However, he explains how non-technical analysts can engage with data in a conversational manner, gaining insights and experimenting with how they ask questions and exploring correlations without the need for advanced technical skills. This democratization of data analysis opens up new possibilities for decision-making and problem-solving. Investors, corporate executives, and researchers can leverage chat GPT to uncover hidden patterns and trends within their datasets. By understanding the correlations between different variables, they can make more informed decisions and develop strategies based on data-driven insights.

The Role of AI Tools in Enterprise Data Analytics

When discussing the use of AI tools at the enterprise level, Adam acknowledges the need for caution and data security. He advises against randomly uploading corporate data into AI tools and highlights the risks associated with data leakage and potential misuse. To address these concerns, he mentions solutions like Microsoft Azure's OpenAI service, which allows organizations to run AI models locally and keep their proprietary data secure. He also mentions Chat GPT’s incognito mode, and the upcoming release of ChatGPT for enterprise tool, which will probably have additional safety guarantees. He talks about what the tool is being used for today such as crunching numbers and making predictions, in addition to coding and analytics and generative AI.

Implications and Forecasting

As the conversation draws to a close, Adam talks about using the tool for forecasting but that it will become better when the technology merges with browsers. He emphasizes the importance of continuous learning and experimentation, as well as the potential for individuals to enhance their skills in domain knowledge, statistics, and technical/data knowledge. He highlights the role of AI tools as a means of human augmentation, assisting users in their data analysis tasks, and talks about his writing and teaching work, and writes about how generative AI is used in teaching and learning. 

Looking ahead, Adam predicts that AI tools will continue to evolve and improve, becoming more user-friendly and capable of handling complex analytics tasks. He emphasizes the need for organizations to embrace these tools while ensuring data security and compliance. By leveraging AI tools effectively, organizations can unlock the full potential of their data and drive better decision-making.

In conclusion, AI-powered tools like Code Interpreter and ChatGPT are revolutionizing the field of data analytics. While they have their limitations, they offer immense potential for organizations and individuals to gain insights from their data. By understanding the capabilities and limitations of these tools, experimenting with different modalities of interaction, and prioritizing data security, organizations can harness the power of AI to drive better analyses, generate value, and make informed decisions in an increasingly data-driven world.

Timestamps:

01:37 Options for uploading data into chat GPT

08:40 The interface of chat GPT code interpreter

12:25 The potential for non-technical analysts to use these tools

13:37 Example of using code interpreter to analyze credit card spending data

15:46 Using code interpreter

21:07 Experimenting with code interpreter and learning Python programming

23:34 Code interpreter can graph data, but limitations exist

25:16 Recommendations for using code interpreter effectively

34:33 Enterprise solutions for using code interpreter with proprietary data

35:45 Current use cases of code interpreter in companies

36:51  Using the GPT-3 tool for forecasting

Links:

Website: https://braff.co/genai-1

Unleashed is produced by Umbrex, which has a mission of connecting independent management consultants with one another, creating opportunities for members to meet, build relationships, and share lessons learned. Learn more at www.umbrex.com.