In this article, I am bringing you the most important AI news for finance professionals of May.

And this month, a lot happened with multiple announcements from the AI Giants!

In this Digest you will learn about:

1. Introduction of GPT-4o
2. Build Your Copilots with Agent Capabilities in Microsoft Copilot Studio
3. Advancements in ChatGPT Data Analysis
4. Google’s Announcements on AI
5. Perform Financial Statement Analysis with LLMs

1. Introduction of GPT-4o

OpenAI has launched GPT-4o (“o” for “omni”), a new flagship model that integrates multiple modalities—text, audio, vision, and video—into a single framework, allowing real-time processing across these inputs and outputs. This model marks a significant step forward in human-computer interaction by supporting a variety of inputs and outputs, including the ability to respond to audio inputs almost as quickly as human response times.

But the biggest news: it’s free, and it’s better than ChatGPT 4!

GPT-4o builds on the capabilities of previous models by offering improvements in non-English language processing, vision, and audio understanding. This enhancement makes it not only faster but also 50% cheaper in the API compared to earlier versions. The model is designed for more natural and efficient user interactions across different formats, whether it’s handling straightforward tasks or engaging in complex conversations.
​
When using GPT-4o, ChatGPT Free users will now have access to features such as:

• Experience GPT-4 level intelligence
• Get responses from both the model and the web
• Analyze data and create charts
• Chat about photos you take
• Upload files for assistance summarizing, writing, or analyzing
• Discover and use GPTs and the GPT Store
• Build a more helpful experience with Memory

GPT-4o also introduces efficiency improvements that make it accessible at lower costs and higher speeds, promising broader availability and usability.

I have asked companies using GPT4 until now, they told me that they almost immediately switched to GPT4o as it’s much cheaper and has better results!

My take:

For finance teams, the real-time processing capabilities of GPT-4o mean quicker access to financial data interpretation and analysis. The model’s ability to handle complex queries across multiple languages and formats.

Moreover, ChatGPT-4o can connect to your Google and Microsoft Drive and generate visual representations like charts and graphs (still bear in mind the confidential part!).

And what I really like: now GPTs, Advanced Analytics and Dall-E are free for everybody!

Interesting articles on the topic:

• Hello GPT-4o​ by OpenAI
  ​
• Introducing GPT-4o​ by OpenAI YouTube

2. Build Your Copilots with Agent Capabilities in Microsoft Copilot Studio

At Microsoft Build 2024, Microsoft introduced new capabilities for Copilot Studio, allowing users to create custom copilots capable of acting as independent agents. These copilots can autonomously handle complex, long-running business processes by being triggered by events rather than just conversations.

This advancement enables applications such as automated IT support, streamlined employee onboarding, and personalized customer service, enhancing productivity and reducing the need for human intervention in routine tasks.

With Copilot Studio, creating and testing copilots is now easier!

You can create your copilot with their brand new conversationally driven experience—describe what you want it to do and what knowledge you want it to have, and Copilot Studio will create your very own copilot (similar to GPTs).

You can then immediately test it out, add additional capabilities, such as your own actions, APIs, and enterprise knowledge—and then publish it live with a few clicks.

My take:

Creating copilots that act as independent agents and that can manage long-running, complex tasks autonomously, will streamline many finance-related processes such as budget tracking, financial reporting, and compliance checks.

This reduces the need for manual oversight and frees up finance teams to focus on strategic analysis and decision-making.

By integrating with enterprise data, these AI copilots can provide real-time insights and proactive management, increasing overall efficiency and accuracy of financial operations and procedures.

Interesting articles on the topic:

• Building your own copilot with Copilot Studio​ by Microsoft
  ​
• Microsoft Copilot Studio: Building copilots with agent capabilities​ by Microsoft

And don’t forget the article we wrote on it last month (we just updated it): https://aifinanceclub.wpenginepowered.com/how-to-create-your-own-finance-copilot/​

3. Advancements in ChatGPT Data Analysis

ChatGPT is rolling out enhancements to its data analysis capabilities, making it more intuitive and user-friendly for Plus, Team, and Enterprise users. The update incorporates the ability to directly upload files from Google Drive and Microsoft OneDrive, facilitating faster access to documents, spreadsheets, and presentations.

This feature streamlines data analysis through an expandable interactive table where users can view and engage with their data in real-time. Additionally, users can customize and interact with various types of charts such as bar, line, pie, and scatter plots, enhancing presentations and documents with insights derived directly from their data.

These improvements build on ChatGPT’s existing capabilities to execute complex data tasks like merging datasets, generating pivot tables, and more, all while ensuring stringent data privacy and security measures are upheld.

ChatGPT is rolling out enhancements to its data analysis capabilities, making it more intuitive and user-friendly for Plus, Team, and Enterprise users. The update incorporates the ability to directly upload files from Google Drive and Microsoft OneDrive, facilitating faster access to documents, spreadsheets, and presentations.

This feature streamlines data analysis through an expandable interactive table where users can view and engage with their data in real-time. Additionally, users can customize and interact with various types of charts such as bar, line, pie, and scatter plots, enhancing presentations and documents with insights derived directly from their data.

These improvements build on ChatGPT’s existing capabilities to execute complex data tasks like merging datasets, generating pivot tables, and more, all while ensuring stringent data privacy and security measures are upheld.

My take:​

The integration of direct uploads from Google Drive and Microsoft OneDrive enables financial analysts to streamline their data workflows by directly importing financial spreadsheets and documents into ChatGPT for analysis.

If your company signed up for a contract with OpenAI and allows you to analyse data with it, you can do a lot of the preliminary work with ChatGPT instead of having to do it manually.

I also think this is a good preliminary view of a feature that will come to all the BI, FP&A and ERP tools: preliminary analytics done for you.

But it won’t replace your specific analysis, closing processes and ad-hoc analysis. Instead it will make you faster and doing the elementary and obvious analysis.

Important as well: the structure of your data becomes more and more important. An unstructured file will be hard to get analyzed inside this tool or Copilot. AI prefers structured data!!

Interesting article on the topic:

• Improvements to data analysis in ChatGPT​ by OpenAI

4. Google’s Announcements on AI

At Google I/O 2024, the “Gemini Era” was introduced, emphasizing advancements in Google’s AI capabilities, particularly in multimodal and long-context processing with Gemini models.

These models integrate various inputs like text, images, and videos, enhancing products across Google’s ecosystem such as Search, Photos, and Workspace.

Key developments include expanding Gemini’s token capacity and integrating it more deeply into consumer applications, promising significant improvements in user interaction and accessibility across languages and formats.

Some of the highlights of yesterday’s Google’s announcements on AI:

• Gemini 1.5 Pro – Their response to GPT4 with 2 million tokens
• Google Search – Now it has an AI agent built in to do complex searches
• Veo – An AI video generator (1080p, more than a minute)
• Imagen 3 – Their response to DALLE – Creates Images
• Project Astra – It aims to help in daily lives with AI

My take:

For all of you who are Google users, this is good news. However, as GPT4o arrived just before this announcement, Google shows that it is still behind in innovation.

What is the most important change is actually on Google Search… but this is less impactful for us in Finance.

As I often say, choosing LLMs is similar to choosing which car to take when traveling between two cities by car. You first need to know how to drive!
With the big LLMs from AI Giants, results are equivalent for 99% of the use cases. The most important thing is knowing how to use them and driving them to your desired outcome.

Interesting articles and links on the topic:

• Google I/O 2024: An I/O for a new generation​ by Google
  ​
• AI Galore: Key Announcements From Google I/O 2024​ by Spiceworks

5. Perform Financial Statement Analysis with LLMs

Last but not least, I want to share a research paper that is not typical news but offers valuable insights into how an LLM can perform financial statement analysis.

Ivo Hop and Christian Martinez also shared this in our Slack Group (thanks guys!)

Key highlights include:

• Performance of LLMs: LLMs were found to outperform financial analysts in predicting earnings changes, demonstrating their potential to provide valuable insights in financial analysis
• Prediction Accuracy: The LLMs’ predictions showed higher accuracy and yielded better investment strategies than traditional models, resulting in higher Sharpe ratios and alphas. Moreover, the prediction accuracy of the LLM matches that of a narrowly trained state-of-the-art ML model.
• Capability without Context: Interestingly, LLMs performed effectively without narrative context, relying solely on the numerical data from financial statements.
• Comparison with Machine Learning Models: LLMs also compared favorably against specialized machine learning models, indicating that their general knowledge and reasoning capabilities are valuable even in highly specialized tasks.

My take:

GPT-4’s ability to outperform human analysts in predicting future earnings, even without narrative or industry-specific context, is particularly impressive.

This is seriously surprising to me, as a few months ago, LLMs had a hard time with computations!

Another sign that we need to look at these models constantly and how well they improve on the computational part, especially as it has been a focus over the last months with more functionalities on data analysis.

​Click here to download the research PDF

That’s it for this intense month!

If you come across any other interesting articles, feel free to share them with the rest of the members in our Slack channel.

I hope you learned new things from this new digest and that this helped you cut the relevant information through the noise!