In this article, I’m sharing the most significant AI news for finance professionals from June.

This month was eventful, with several announcements from major AI companies!

In this digest, you will learn about:

1. The Arrival of Claude 3.5 Sonnet as a Serious Competitor to ChatGPT 4o
2. How to use Prompt Engineering to Reduce Hallucinations
3. The New Power BI integration with Copilot
4. SAP Joule’s latest integration with Microsoft Copilot
5. How two of the world’s biggest firms (Amazon and Goldman Sachs) are moving more towards the use of Generative AI

1. The Arrival of Claude 3.5 Sonnet as a Serious Competitor to ChatGPT 4o

Anthropic’s latest release, Claude 3.5, codenamed “Sonnet,” marks a significant leap in AI capabilities, directly challenging industry giants like OpenAI’s GPT-4o.

Based on the demonstrations of experts, Claude 3.5 is at par in terms of performance and offers some practical features especially the “Artifact” view.

What is an artifact? When you prompt the chatbot to produce content such as code snippets, text documents, games, or website designs, these Artifacts will be displayed in a separate window next to their conversation showing immediately the result.

My take:

For me, the number one revolution is to have the artifact feature. You can see here a video of Claude 3.5 with the artifact feature in action.

Here is a football game I coded in one prompt:

I have checked several comparisons and Claude 3.5 seems to have an edge on coding but is slower to produce the output compared to ChatGPT 4o.

And for the Europeans, good news: Claude is since May 2024 available in Europe! So you don’t need a VPN anymore 😉

Interesting articles on the topic:

• Anthropic’s Claude: What To Know About ChatGPT Rival After Latest Model Trounces Industry Giants​ by Forbes
  ​
• ​Claude 3.5 Sonnet: The Future of AI Collaboration?​ by CMSWire

2. How to use Prompt Engineering to Reduce Hallucinations

We have found several articles showing which Prompt Engineering techniques to use to reduce hallucinations.

For example, the Chain-of-Verification (CoVe) technique:
Chain-of-verification (CoVe) prompts explicitly require the model to provide step-by-step verification for its responses by citing external authoritative sources.

Another one is to require to cite its sources when mentioning facts or asking where can the facts could be proof checked.

One innovative technique is also the Chain-of-knowledge (CoK). When using it, you explicitly require the model to source its responses from chains of expert knowledge to reduce logical leaps or false inferences.

These methods not only minimizes the chances of generating incorrect or irrelevant information but also boosts the model’s ability to maintain context throughout interactions.

OpenAI is also testing a tool to proof-check code which is called “CriticGPT”. Ultimately, the model will build mechanisms to fact and proof-check the responses in order to minimize the work of the user.

My take:

In my training, I often get asked how to verify the accuracy of the output of the LLM models.

I demonstrate some of these techniques, which you can discover in the articles below or the ones that I teach in my Prompt Engineering Course.

Interesting articles on the topic:

• Improve Accuracy and Reduce Hallucinations with a Simple Prompting Technique​ by PromptHub
  ​
• ​Finding GPT-4’s mistakes with GPT-4 by OpenAI
  ​
• “Advanced Prompt Engineering for Reducing Hallucination​” by Bijit Ghosh on Medium

3. New Power BI integration with Copilot

Microsoft has introduced the Power BI Copilot, a feature designed to enhance user experience by allowing users to interact with data through natural language queries. This tool is currently in public preview and aims to simplify data analysis by leveraging AI capabilities.

Users can ask questions about their data directly within the Power BI interface, and the Copilot will provide relevant insights or visualizations. This approach is intended to make data more accessible to users who may not have advanced technical skills.

Additionally, Microsoft has provided detailed guidance on how to use the Copilot effectively, including best practices for asking questions and examples of common queries.

My take:

I believe that the introduction of Power BI Copilot will help non experience users to create dashboards and reports in a faster way.

For me Power BI was already the most intuitive Business Intelligence and Data Visualisation tool.

By adding the Copilot functionality, it makes it even more accessible for finance teams that want to build faster and better dashboards or reporting for their business partners.

One caveat, as for Copilot in Excel or Powerpoint, the use of Copilot in Power BI is still at its infancy and an experienced user might get frustrated as this is for now really simplistic.

Interesting article on the topic:

• Ask Copilot questions against your semantic model (Preview)​ by Microsoft
  ​
• Copilot Ask Data Questions by Microsoft

4. SAP Joule’s latest integration with Microsoft Copilot

SAP and Microsoft have announced a partnership integrating SAP’s Joule AI with Microsoft’s Copilot, creating a unified work experience.

This collaboration aims to streamline workflows by embedding AI capabilities into business applications. Joule, with its natural language capabilities, allows users to interact more intuitively with SAP systems, while Microsoft’s Copilot provides integration with tools like Microsoft 365.

This partnership also includes enhanced data analytics and security features.

In the demonstration from SAP, you can see how you can pull data from the ERP by just chatting in Teams with SAP Joule.

There is no exact release date communicated, but here is what SAP is communicating on its website:
​“SAP and Microsoft will start integrating their Joule and Copilot for Microsoft 365 later this year, combining enterprise data residing in SAP with contextual knowledge from Microsoft 365, including Microsoft Teams, Microsoft Outlook, Microsoft Word, and more, delivering richer insights for better decision-making. “

My take:

I have worked for more than 8 years with SAP and it’s old interface can be frustrating because it requires specific knowledge of transactions and specific parameters.

Simplifying user interaction will definitely empower SAP users to achieve more. I also think allowing users to interact with SAP in the Office environment is a smart partnership for both companies. But unfortunately, we don’t have a release date yet (they announced “end of the year”).

Check the examples in the video linked below to see how it will look like.

Interesting articles and links on the topic:

• Joule to Integrate with Microsoft Copilot for a Unified Work Experience​ by GoogSAP News Center
  ​
• Innovation Keynote Highlights | SAP Sapphire 2024​ by SAP

5. How two of the world’s biggest firms (Amazon and Goldman Sachs) are moving more towards the use of Generative AI

Amazon and Goldman Sachs are both leveraging generative AI to enhance operational efficiency and decision-making.

Amazon has introduced Amazon Q, a generative AI assistant integrated into its AWS platform. This tool aims to streamline financial management by enabling employees to access data from various sources and generate insights quickly.

Amazon Q allows users to create custom applications, build business intelligence dashboards, and automate routine tasks, ultimately improving productivity and data-driven decision-making across departments.

Meanwhile, Goldman Sachs has launched its first generative AI platform: the GS AI Platform. Select employees can use the platform for custom applications, such as a copilot tool for investment bankers to search and analyze documents.

The first tool they launched is Microsoft’s GitHub Copilot, which is expected to boost developers’ efficiency by 20%. Additional AI tools for document translation and research summarization are in the early stages.

My take:

From my perspective, it’s always interesting to see real use cases from business giants.

For example, the Amazon tax compliance team created a tool that uses generative AI to help validate inbound invoices for VAT.

For Goldman Sachs, we can see that they are not only relying on one model but on the combination of different ones (GPT, Gemini, Llama) and are enriching it with their own proprietary data.

I believe for big companies like Goldman, fine-tuning existing model with their own data is the right strategy, as the value will come your proprietary data and that’s where you should spend time, rather than building your own LLM model.

Interesting articles and links on the topic:

• Amazon Leans Into Generative AI to Manage Its Finances​ by WSJ
  ​
• Goldman Sachs Deploys Its First Generative AI Tool Across the Firm​ by WSJ

That wraps up an exciting month!

If you find any other intriguing articles or insights, please share them with everyone in our Slack channel.

I hope this digest helped you discover valuable information amidst the noise and that you learned something new along the way.

Let’s keep the conversation going!