English for Special Purposes

AI Development English

High-level technical English for AI engineers, researchers, product managers, data specialists, and safety teams.

8 modules
42 field terms
Interactive practice

Printable Curriculum

Download the full materials

Instructor Guide Participant Workbook Dialogue Lab Jargon Guide

Web Practice Lab

Practice the decisions, not only the vocabulary

Use the activities below to rehearse how a professional in this field clarifies risk, pushes back, and turns pressure into a concrete next step.

Module Focus

Scenario Coach

Respond under pressure

Your professional response

Jargon Flashcard

Pushback Builder

Build a four-step response

Evidence or fact Risk or consequence Decision or next step

Dialogue Coach

Model line

Language notes

Progress

Practice checklist

I used at least two field terms accurately. I named the evidence or policy boundary. I gave a concrete next step. I avoided overpromising.

0 of 4 complete

Student PDF in Web Form

Module map

Open Participant Workbook PDF

Speaking the AI Development Stack

AI teams use a layered vocabulary: model, data, prompt, retrieval, tools, serving, evaluation, monitoring, and product experience. Learners need to locate a problem in the stack before they can discuss it clearly.

LLM, Transformer, Parameter, Checkpoint

LLMs, Transformers, Tokens, and Context

High-level AI communication often depends on explaining what the model sees: tokens, messages, context window, instructions, examples, retrieved text, and tool results.

Foundation model, Multimodal, Prompt, System prompt

Data, Datasets, Labels, and Leakage

AI systems are shaped by data quality. Teams need precise language for dataset splits, annotation guidelines, leakage, imbalance, representativeness, and privacy constraints.

Few-shot, Context window, Token, Temperature

Retrieval, Embeddings, Vector Search, and RAG

Many production AI apps combine retrieval with generation. Learners need to discuss chunking, embeddings, vector stores, recall, reranking, grounding, citations, and retrieval misses.

Embedding, Vector store, Chunking, Reranker

Fine-Tuning, Alignment, and Adaptation

Teams often confuse prompt changes, RAG, fine-tuning, adapters, supervised fine-tuning, preference tuning, and RLHF. The language goal is to recommend the right adaptation method for the problem.

RAG, Grounding, Fine-tuning, SFT

Evaluation, Benchmarks, and Regression

AI teams need language for uncertainty. 'It looks better' is not enough. Learners need to discuss offline evals, online evals, golden sets, human review, model-graded evals, regression, pass rate, and confidence.

RLHF, DPO, LoRA, Adapter

Inference, Latency, Cost, and Deployment

AI development is also systems engineering. Learners need vocabulary for inference paths, throughput, batching, caching, rate limits, GPUs, quantization, streaming, fallbacks, and SLOs.

Eval, Benchmark, Golden set, Regression

Safety, Security, Privacy, and Governance

AI teams must discuss risk precisely: hallucination, prompt injection, jailbreaks, PII, data retention, bias, harmful output, policy enforcement, audit logs, and human-in-the-loop review.

Pass rate, LLM-as-judge, Inference, Latency

More EFSP Tracks