| Date | Topic | Description | Events | Deadlines |
|---|---|---|---|---|
| 9/22 | Introduction | LLMs hallucinate. How to turn LLMs into useful, dependable, general conversational assistants? Concept of computational thinking. Carefully designed LLM-based algorithms for reading, writing, retrieving data; formal representation of knowledge and conversational policies. Course design and outline. | ||
| 9/24 | Knowledge Curation | How to use LLMs to perform research in an open domain? Research in pre-writing stage by iteratively searching and reading from different perspectives; adding interactivity to allow personalization. System: Storm/Co-Storm | Homework 1 out | Student intro form due |
| 9/29 | Building a task-oriented agent | How to create an agent that does not hallucinate? Genie Worksheet is the first high-level task-oriented agent specification language that lets users focus on the task to be done; low-level dialogue implementation details are left to the implementation of the language. How to scale the creation of effective and reliable agents across different domains easily? Implementation of the Genie Worksheet; formal dialogue state representation; semantic parsing; dialogue state tracking; response generation. | ||
| 10/1 | Research project ideas | What are the ongoing research projects that students can participate in? The major themes: Creating a deep researcher: Acquisition of new data. Data analysis on databases, long documents, literature. Formal semantic representations: Extracting Satisfiability Modulo Theories (SMT), Knowledge Graphs (KG), Rule-based Conversational Policies. Real-life applications from cancer drug resistance research, sustainability modeling, clinical trial matching, journalism. | Project Intent Out | Homework 1 due on 10/02 |
| 10/6 | Evaluation of Task-Oriented Agents |
Comparisons between different conversational agent architectures; Evaluation of task-oriented agents. | Homework 2 Out | |
| 10/8 | Student project ideas | Students pitching preliminary project ideas | Project Proposal Assignment out | Project Intent due |
| 10/13 | Grounding Conversational Agents on Free Text | How to create a hallucination-free conversational bot grounded on free-text? RAG Text retrieval; Summarization; Verifying generation; Response generation; Evaluation methodology. System: WikiChat | Homework 2 due on 10/13 | |
| 10/15 | Project Proposals [feedback form] [presentation schedule] |
Groups present their proposals | Project Proposals due | |
| 10/20 | Project Proposals | Groups present their proposals | Weekend Update due 15:00 (gradescope) | |
| 10/22 | Introduction to Agents for Structured and Hybrid Data | How to create a hallucination-free conversational bot grounded on structured data? Semantic parsing; Databases; Expressiveness of database queries; Few-shot prompting on small schemas; Handling enumerated types; Comparison with human annotations. | ||
| 10/27 | Structured / Unstructured Query Language | How to answer questions combining structured and unstructured data? SUQL: The first query language that combines information retrieval (NL) with database queries (SQL); Language design; Evaluation methodology. | Weekend Update due 15:00 (gradescope) | |
| 10/29 | Question Answering on Sets of Long Documents | How to answer questions that require reading, analyzing, and comparing across large documents in a set? Training and Chunk-based Approaches. Challenges of Chunking. SLIDERS is a new document analysis system which uses automatic schematization to convert text into a database tailored to given questions. Techniques include semantic-driven chunking, contextualized extraction, and chunk reconciliation | ||
| 11/3 | Document Set Analysis: Qualitative Coding | How to perform research on a large set of documents? Qualitative coding is a classic manual method used across many fields including social sciences, healthcare, education, business and marketing, and technology. Challenges in automating qualitative coding. Abstractive event detection. | Course project final deliverable details released | Weekend Update due 15:00 (gradescope) |
| 11/5 | Agentic AI for Knowledge Base Queries | How to handle complex knowledge tasks using the agentic approach? E.g. Generating SPARQL query for Wikidata; Action set design; experimental approach | ||
| 11/10 | Satisfying Natural Language Constraints Using SMT | How to determine if a document satisfies complex logic constraints given by another document. Satisfiability modulo theories (SMT). A real case study with clinical trial matching. | Weekend Update due 15:00 (gradescope) | |
| 11/12 | NLP Building blocks | Vision-Language Models to Make Historical Documents Accessible | ||
| 11/17 | Multimodal Applications | How to build a multi-modal app that supports complex commands? Motivation; Arbitrary composition of APIs in a program by voice; Combining graphical and voice outputs; Showing voice command results in native graphical outputs; How to discover features; ReactGenie framework, GenieWizard. | Weekend Update due 15:00 (gradescope) | |
| 11/19 | Training LLMs | How do we create LLMs? instruction following models; training data. | ||
| 11/24 | Thanksgiving Break | |||
| 11/26 | Thanksgiving Break | |||
| 12/1 | No class | Weekend Update due 15:00 (gradescope) | ||
| 12/3 | Final project presentation (3 hours) | Checkout Project Gallery |