Source 1
Effective Prompts for AI: The Essentials
MIT Sloan Teaching & Learning Technologies introduces prompting as 'programming with words' and highlights three practical principles: provide context, be specific, and build on the conversation.
Promptly Research Labs/Papers
Papers & sources
This page collects the MIT and arXiv references used across the research experience. The annotations are intentionally short so you can scan for theory, prompting guidance, stability work, and optimisation methods quickly.
Open source
Source 2
Glossary of Terms: Generative AI Basics
MIT Sloan's glossary gives concise definitions for prompt engineering, hallucinations, context windows, and related terms used throughout the Labs page.
Source 3
The Prompt Report: A Systematic Survey of Prompt Engineering Techniques
A broad survey that organizes prompt engineering into a shared taxonomy and vocabulary, making it useful for explaining prompt patterns, few-shot examples, retrieval grounding, and optimisation loops.
Source 4
Automatic Prompt Optimization with "Gradient Descent" and Beam Search
Introduces Automatic Prompt Optimization, where natural-language critiques act like textual gradients to guide search over rewritten prompts.
Source 5
Measuring and Controlling Instruction (In)Stability in Language Model Dialogs
Measures instruction drift in multi-turn dialog and links part of the effect to attention decay, which is directly relevant when explaining why prompt structure can matter across long chats.
Source 6
A Theoretical Framework for Prompt Engineering: Approximating Smooth Functions with Transformer Prompts
A recent theory paper with a Harvard-affiliated author that frames prompts as inference-time configurations capable of shaping transformer computation.
Source 7
Large Language Models Are Human-Level Prompt Engineers
Presents Automatic Prompt Engineer (APE), which treats instruction search as a program-synthesis style optimisation problem over candidate prompts.
Source 8
Large Language Models as Optimizers
Explores optimisation by prompting (OPRO), where an LLM iteratively proposes new candidates after seeing prior solutions and scores.