Frequently Asked Questions

Course Logistics

Who is this course for?

This is a PhD-level course designed for doctoral students and researchers interested in building autonomous AI systems. It's also suitable for advanced Master's students and industry practitioners with strong programming skills.

What are the prerequisites?
  • Strong Python programming skills
  • Familiarity with machine learning concepts
  • Basic understanding of transformer architectures
  • Ability to read and understand research papers
How much does API access cost?

Most exercises can be completed with $10-20 of API credits. Both OpenAI and Anthropic offer free credits for new accounts. We also use gpt-4o-mini which is very cost-effective.

Can I audit the course?

All materials are freely available online. You can follow along with lectures and notebooks at your own pace. For official credit, you must be enrolled at the institution.

Technical Setup

What Python version do I need?

Python 3.11 or higher is recommended. Some features may work with Python 3.10, but 3.11+ is preferred for best compatibility.

Do I need a GPU?

No, all exercises use API-based models. A standard laptop is sufficient. For local model experiments (optional), a GPU would be helpful but not required.

Can I use Anthropic instead of OpenAI?

Yes! Most notebooks support both providers. Claude models work well for agent tasks. Check the specific notebook for provider options.

How do I set up my API keys?

Create a .env file in the project root with:

OPENAI_API_KEY=sk-your-key-here
ANTHROPIC_API_KEY=sk-ant-your-key-here

Never commit this file to version control!

Course Content

What's the difference between an agent and a chatbot?

A chatbot responds to queries in a single turn. An agent can take multiple actions, use tools, maintain state across interactions, and pursue goals autonomously. Agents have a reasoning loop while chatbots are typically stateless.

Why do we need agent frameworks?

Frameworks like LangGraph handle common patterns: state management, error handling, tool orchestration, and multi-step workflows. They save development time and provide tested abstractions.

How do I prevent my agent from hallucinating?

See Week 9 for detailed techniques. Key strategies include: grounding responses in retrieved facts, using verification loops, implementing claim checking, and expressing uncertainty appropriately.

Which agent framework should I use?

It depends on your needs:

  • LangGraph: Best for complex, stateful workflows
  • CrewAI: Best for role-based team collaboration
  • AutoGen: Best for conversational multi-agent systems
  • Plain Python: Best for learning and simple agents

Assignments

Can I use LLM assistants for assignments?

Yes, but you must disclose their use. The goal is learning - using an LLM to write all your code defeats the purpose. Use them for debugging, explanation, and exploration.

How are assignments graded?

Assignments are evaluated on:

  • Correctness of implementation
  • Code quality and documentation
  • Understanding demonstrated in comments/reports
  • Creativity in extensions (bonus)
Can I collaborate with others?

Discussion is encouraged, but submissions must be individual. Pair programming on understanding is fine; copying code is not.

Research

What are good research directions in agentic AI?

Hot topics include:

  • Long-horizon planning with uncertainty
  • Grounded world models for agents
  • Agent safety and alignment
  • Efficient tool learning
  • Multi-agent coordination at scale
  • Evaluation methodologies
How do I find a research topic?
  1. Read recent papers from top venues (NeurIPS, ICML, ACL)
  2. Identify limitations mentioned in papers
  3. Look for gaps between theory and practice
  4. Consider applications in your domain
  5. Discuss with the instructor

Getting Help

Where can I ask questions?
I'm stuck on a notebook. What should I do?
  1. Read the error message carefully
  2. Check the troubleshooting section in Getting Started
  3. Search for similar issues on GitHub
  4. Post on GitHub Discussions with code and error
  5. Attend office hours

Don’t see your question? Ask on GitHub Discussions!

Back to top