# AI Foundations Track Overview Independent preparation for the OpenAI Academy AI Foundations course – seven domains, a decision-first mindset, hands-on ChatGPT exercises and two independent mock exams. import { Card, CardGrid, Badge } from '@prosefly/astro-components'; Track OAI-FOUND 7 domains 2 × 60-item mocks Academy-aligned This is the entry track. It assumes no prior AI experience and builds, from first principles, the working knowledge a knowledge worker needs to use ChatGPT well and judge its output. If you are new to AI, start here; every other OpenAI track on this site assumes what this one teaches. ## What this track prepares you for This track mirrors the OpenAI Academy course **AI Foundations** – a 60–75 minute, ChatGPT-based course and the first course in the **Foundations pathway** (AI Foundations → Applied AI Foundations → Agents and Workflows). The Academy course covers AI basics, prompting, context and responsible use; this track goes deeper on each so that recall and judgment survive a randomised assessment. To be precise about what a pass earns you: completing the Academy course and scoring at least 80% on its assessment earns an **OpenAI Academy badge** issued through Accredible. That badge, and any pathway certificate of completion, is **not a certification** and does not guarantee eligibility for a future OpenAI certification – that is OpenAI's own wording. The mock exams here are **independent practice**, built from publicly available learning objectives; they are not official OpenAI questions and there is no public OpenAI exam to have questions from. See the [credential landscape](/openai/credentials/) and [assessment model](/openai/assessment-model/) pages for the full picture. ## Blueprint | # | Domain | Weight | Items (of 60) | Track page | | --- | --- | --- | --- | --- | | 1 | AI and Generative AI Fundamentals | 14% | ~8 | [Domain 1](/openai/foundations/domains/d1-ai-and-generative-ai-fundamentals/) | | 2 | How Language Models Behave | **16%** | ~10 | [Domain 2](/openai/foundations/domains/d2-how-llms-behave/) | | 3 | ChatGPT Surfaces and Features | 14% | ~8 | [Domain 3](/openai/foundations/domains/d3-chatgpt-surfaces-and-features/) | | 4 | Prompting and Instructions | **20%** | ~12 | [Domain 4](/openai/foundations/domains/d4-prompting-and-instructions/) | | 5 | Context, Files and Memory | 12% | ~7 | [Domain 5](/openai/foundations/domains/d5-context-and-files/) | | 6 | Verifying and Evaluating Output | 14% | ~8 | [Domain 6](/openai/foundations/domains/d6-verifying-and-evaluating-output/) | | 7 | Responsible and Safe Use | 10% | ~6 | [Domain 7](/openai/foundations/domains/d7-responsible-and-safe-use/) | :::tip[Where the marks are] Prompting (20%), How Language Models Behave (16%) and the two evaluation-flavoured domains – ChatGPT Surfaces (14%) and Verifying Output (14%) – together are **64%** of the mock. This track rewards knowing *why* the model behaves as it does and *how* to steer and check it, far more than it rewards memorising product names. ::: :::note[Two independent mock exams] This track ships **two** full-length, domain-weighted independent mock exams of 60 items each. [Mock Exam 1](/openai/foundations/practice-exam/) is your diagnostic – sit it first, untimed, to find your two weakest domains. [Mock Exam 2](/openai/foundations/practice-exam-2/) is deliberately harder (more multi-constraint stems and `FIRST` / `BEST` / `MOST cost-effective` / `TWO` qualifiers) and timed – use it as your readiness gate before you enrol in and sit the real Academy assessment. All items across both mocks and the domain pages are distinct. ::: ## The mindset this track rewards The Academy course frames ChatGPT as tutor, practice environment and feedback loop at once. The assessment rewards one posture throughout: **ChatGPT is a fast, fluent, fallible collaborator whose output you own.** Correct answers consistently: - Treat fluent, confident text as a draft to verify, not a fact to forward. - Give the model the context it needs instead of expecting it to guess. - Match the effort and the surface to the stakes – a quick reply for a low-stakes task, reasoning and verification for a consequential one. - Keep sensitive data out of tools and conversations that are not sanctioned for it. - Route irreversible, regulated or external work through a human review gate. Wrong answers trust the output because it reads well, paste confidential data without checking policy, expect one giant prompt to do everything, or treat the model as a search engine that cannot be wrong. ## Suggested time allocation A 12–15 hour plan, weighted by `weight × how new the material is to you`. | Domain | Weight | Hours | | --- | --- | --- | | Prompting and Instructions | 20% | 3 | | How Language Models Behave | 16% | 2.5 | | AI and Generative AI Fundamentals | 14% | 2 | | ChatGPT Surfaces and Features | 14% | 2 | | Verifying and Evaluating Output | 14% | 2 | | Context, Files and Memory | 12% | 1.5 | | Responsible and Safe Use | 10% | 1.5 | ## Hands-on preparation checklist Do these in a real ChatGPT account. Reading about prompting teaches you nothing about prompting. - [ ] Ask ChatGPT the same factual question three times in new chats and note where the wording of the answer differs – this is non-determinism in the raw. - [ ] Take one weak prompt and rebuild it with role, task, context, constraints, format and a success criterion; compare the two outputs side by side. - [ ] Ask for an obscure statistic with no source, then ask ChatGPT to cite it – observe how a hallucinated citation looks. - [ ] Switch between a fast model and a reasoning model on a multi-step problem and compare quality, latency and the visible reasoning. - [ ] Create a **Project** with custom instructions and one uploaded file, then ask a question that only the file can answer. - [ ] Turn a source document into a `Search` query and a `deep research` request and compare what each returns. - [ ] Add a fact to **memory**, start a new chat, and confirm the model uses it; then find and delete that memory. - [ ] Draft an external customer email, then write the human-review checklist you would apply before sending it. ## Track pages 14% 16% 14% 20% 12% 14% 10% 60 items · diagnostic 60 items · harder, timed