Hello there,
Last week, I spoke with a financial professional frustrated by AI-generated misinformation during critical tasks.
"We can't afford wrong information," he told me. "Our reputation depends on accuracy."
Does this sound familiar?
AI hallucinations (those confidently wrong answers) happen more often than you'd like.
Here’s how to avoid them.
The Core: 4 Steps to Prevent AI Hallucinations
1. Feed Your Own Source Data (RAG Strategy)
Retrieval-Augmented Generation (RAG) connects ChatGPT directly to your internal databases or data sources like Google Drive or SharePoint.
Result: Your AI’s responses stay grounded in your verified documents, not its imagination.
2. Lower the Temperature
Set your model’s temperature (a parameter that controls the randomness and creativity of the model's output) between 0 and 0.3 to force more precise and deterministic outputs.
Adjust this setting via API calls or enterprise dashboards labeled “Precise Mode.”
3. Ask for Citations
Explicitly request citations in your prompts: “Cite every claim with a URL or document title.”
Determine that critical statements are backed by verifiable sources. No citation? Flag it immediately for review.
4. Build a Human Review Loop
Implement a KPI: Any AI output over 250 words requires a human review.
Result: Confidence in AI outputs skyrockets, saving time, and reducing costly mistakes.
AI News That Matters for Finance Pros (May 20th – May 27th)
AI continues to reshape our financial landscape and redefine what's possible. This week's roundup captures the most impactful and intriguing developments. I’ve handpicked these articles to keep you informed, inspired, and ahead of conversations happening right now in finance and beyond.
Citi Introduces AI Tools for Employees in Hong Kong
Citigroup has launched "Citi AI," a suite of AI tools designed to assist employees in tasks such as retrieving policy information, summarizing documents, and drafting communications. This initiative aims to streamline internal operations and aligns with Hong Kong's push for responsible AI adoption in banking. Read more
Go Fintech Plans Australian Expansion with AI-Driven Inclusive Finance
Go Fintech, a leader in AI-powered financial technology, is set to expand into the Australian market. The company aims to promote inclusive finance by leveraging AI to enhance accessibility and effectiveness in financial services. Read more
OpenAI Acquires Jony Ive’s Startup to Develop AI Companions
OpenAI has acquired Jony Ive's startup, io, for $6.5 billion, with plans to develop AI "companions." This move signifies a significant step towards integrating AI into daily life through personalized, interactive agents. Read more
Google Unveils Veo 3: AI Video Model with Audio Generation
Google has introduced Veo 3, an advanced AI model capable of generating videos with synchronized audio. This development pushes the boundaries of AI-generated multimedia content. Read more
AI Agents Pose Emerging Security Risks, Study Finds
A study by Dimensional Research and SailPoint reveals that 23% of IT professionals have witnessed AI agents being tricked into revealing access credentials. The report highlights the need for stringent access controls and oversight as AI agents become more prevalent in enterprises. Read more
AI-Generated Reading List Includes Nonexistent Books
A national summer reading list, partially generated by AI, mistakenly featured 10 fictitious books attributed to real authors. The error underscores the limitations of AI in content curation and the importance of human oversight. Read more
Call to Action
📩 What's your biggest challenge with AI at work?
Hit reply and let me know—I might feature your story anonymously next week.
Kind regards,
Derwish Rosalia
Certified Accountant & AI Expert
15+ Years in Big 4, Corporate, and Government Roles