References: Beyond the chat window
Source material
Section titled “Source material”This lesson is an original adaptation. Its structure, classroom demos, and quoted lines come from the fifth class of a Harvard Kennedy School course; the prose, framing, and examples for Clawdemy readers are our own. Clawdemy is independent of Harvard, which has not reviewed or endorsed this track.
- The Science and Implications of Generative AI (HKS DPI-681M), Harvard Kennedy School, Spring 2024. Faculty: Sharad Goel, Dan Levy, and Teddy Svoronos. This lesson adapts Class 5: Beyond chatbots from the Spring 2024 course site, whose content is licensed under Creative Commons Attribution 4.0. The system prompt demos, the admissions-office activity, the fine-tuning framing, and every quoted line, including “valid for every single interaction,” “typically harder and more expensive,” “something that most of us can do,” “the designer has a lot of control on how it’s going to respond,” and “the logic behind the tools is the same,” come from this class’s lecture videos.
- Official course lecture playlist on YouTube, Harvard Kennedy School. The full lectures, free to watch. We mean it when we encourage you to take the original course alongside this track: good teachers deserve more students.
- Provenance note: quotations and classroom details in this lesson were drawn from the transcripts of the official Class 5 lecture videos in the playlist above, obtained from the official Harvard sources only. No third-party re-uploads or mirrors were used.
The lesson’s mid-2026 claims about voice conversations, that many assistants offer them and that in most tools speech is converted to text, run through the model, and converted back to sound, were verified against current industry engineering sources rather than the 2024 course material:
- Softcery Lab on real-time versus turn-based voice agent architecture, an engineering breakdown of how production voice agents chain speech-to-text, a language model, and text-to-speech, and where newer real-time designs differ.
- AssemblyAI on speech-to-speech and voice agent APIs, a survey of current voice agent architectures, including the pipeline pattern most tools use today and the emerging speech-native alternatives, which is why the lesson hedges with “in most tools.”
Going deeper
Section titled “Going deeper”- Official HKS course listing for DPI-681M, Harvard Kennedy School. The in-person course behind the open online materials this track adapts.
On this site
Section titled “On this site”- Asking well: the anatomy of a good prompt, lesson 2 of this track. The task, instructions, context anatomy that this lesson makes permanent.
- How RAG works: grounding a model in your docs (Transformers and LLMs track, shown as AI Foundations in the sidebar). The machinery underneath the automated librarian, for readers who want to go a level down.
- How prompting works: system prompts, injection (same track). What system prompts look like from the builder’s side, including how they can be attacked.
Adjacent topics
Section titled “Adjacent topics”- The next lesson in this track turns capability into judgment: should AI do this task at all, the course’s decision framework for that question, and a cautionary tale about an official chatbot that gave the public wrong answers.
- The made-up-answers problem that retrieval reduces but does not eliminate gets its own treatment near the track’s end, in the arc on risks and what to do about them.