AuditChats
Our signature virtual community connecting research and practice.
What is AuditChats?
AuditChats is a (mostly) virtual community aimed at connecting researchers and real-world practitioners to discuss professional challenges, develop and disseminate cutting-edge audit research, and work together to solve practice problems.
Connect & Share
Connect real world professionals with researchers to assist researchers with working on relevant research and help professionals consume research findings.
Support Doctoral Students
Assist current auditing doctoral students with networking, obtaining visibility, and securing high-quality job placement.
Foster Collaboration
Foster collaboration among research that transcends school boundaries.
Ready to Join the Conversation?
Event Archive & Recordings Library
Explore our comprehensive library of past AuditChat sessions and recordings.
No AuditChat events available at this time. Please check back soon!
Key Takeaways
Core LLM Mechanics
- • LLMs generate text through probabilistic token prediction, not rules or reasoning
- • Models balance between most probable responses and unique generations using probability distributions
- • Context window determines how much information the model can consider
- • Custom instructions and system prompts shape model behavior behind the scenes
Strategic Prompting Philosophy
- • Having a clear goal and enough knowledge to describe it is more important than perfect prompt syntax
- • Start with simple prompts and iterate rather than seeking perfection initially
- • When hallucinations occur, restart the conversation rather than trying to correct within the same chat
- • Combine multiple prompting strategies for complex tasks
Writing Enhancement Workflow
- • Use LLMs as writing assistants rather than replacements
- • Train models on your writing style through examples
- • Break complex writing tasks into structured data points
- • Implement multi-stage verification including third-party perspective checks
Pedagogical Applications
- • Design interactive learning experiences with guided progression
- • Implement role-reversal where students teach concepts back to the model
- • Create personalized feedback loops with multiple clarifying questions
- • Use GPTs to scale individualized student support
Advanced Implementation
- • Temperature parameter (0-2) controls response creativity vs predictability
- • API integration enables automated processing of large datasets
- • System prompts provide consistent behavior across interactions
- • Custom GPTs can maintain complex instruction sets across sessions
Practical Considerations
- • Balance efficiency gains against accuracy requirements
- • Consider privacy implications when uploading sensitive content
- • Expect initial time investment before seeing productivity gains
- • View LLMs as assistants that complement rather than replace human expertise
Featured Speakers & Topics
Nick Hallman
Associate Professor
API Integration
- • Integrating LLMs with APIs
- • Real-world use cases
- • Precise instructions
Kiryll Pokudin
LLM Expert in Residence
LLM Overview
- • What are language models?
- • How does it work?
- • Hallucinations!
Dorothy Dickmann
PhD Student
Prompting Strategies
- • Strategies for getting better results
- • Iterative refinement
- • Precise instructions
Jaime Schmidt
Professor
Writing Letters
- • Using LLMs for letter writing
- • Iterative refinement
- • Precise instructions
Brent Garza
Clinical Professor
GPTs
- • Using LLMs in educational settings
- • Iterative refinement
- • Precise instructions
Event Resources
Loading transcript...
LLM Overview
Kiryll Pokudin
• What are language models?
• How does it work?
• Understanding hallucinations
API Integration
Nick Hallman
• Integrating LLMs with APIs
• Real-world use cases
• Technical implementation
Prompting Strategies
Dorothy Dickmann
• Strategic prompting techniques
• Iterative refinement
• Best practices
GPTs in Education
Brent Garza
• Educational applications
• Custom GPT creation
• Student engagement
Q: How do you handle LLM hallucinations in critical applications?
A: The key is implementing verification systems. Never rely on a single LLM output for critical decisions. Use techniques like:
- Multiple model validation (ask different models the same question)
- Source verification (require citations and check them)
- Human oversight for final approval
- Starting fresh conversations when errors are detected
Q: What are the best practices for API integration in audit workflows?
A: Start small and scale gradually. Key considerations include:
- Data privacy and confidentiality requirements
- Error handling and fallback procedures
- Rate limiting and cost management
- Regular model version updates and testing
Q: How can academic researchers get started with LLMs?
A: Begin with clear, specific goals rather than trying to learn everything at once:
- Identify one specific task you want to improve
- Start with simple prompts and iterate
- Join communities and share experiences
- Experiment with free tools before investing in APIs
Got More Questions?
Ask them on this topic's discussion board! Join the conversation with other participants and continue the learning.
Join Discussion Board