General-Purpose AI Assistants

What ChatGPT, Claude, Gemini, and the Stanford AI Playground offer, and what common assistant features mean.

About 15 minutes

By the end of this lesson you should be able to

Describe what ChatGPT, Claude, Gemini, and the Stanford AI Playground are, and how Stanford offers each
Explain what common assistant features do, and one thing each feature label does not guarantee
Match a task to the features it needs, including tasks where ordinary software is enough

What an assistant can help with

A general-purpose AI assistant is a chat-based tool you can ask to explain, draft, summarize, brainstorm, or reorganize, on almost any subject, and then question further. Most also bundle features such as web search, file reading, an editing pane, and voice. This lesson explains what those features mean, gives an SLS example of each, and notes what each label does not promise.

Product names, features, and Stanford arrangements below were checked against the providers’ help pages and Stanford’s service pages on September 25, 2026. They change often; the linked pages govern.

Four assistants you can reach through Stanford

Illustrative examples, not tested in an SLS account and not rankings: each assistant can do all of these kinds of work to some degree.

ChatGPT Edu: OpenAI’s ChatGPT, for explaining, drafting, web search, questions about uploaded files, and data analysis. Illustrative use: upload a public report, ask for its main recommendations, then ask for a comparison table that you check against the report. Stanford offers a free Standard account to faculty, students, postdocs, and staff, requested through University IT. (Stanford UIT: OpenAI ChatGPT Edu (opens in a new tab))
Claude for Education: Anthropic’s Claude, for explaining, drafting and critiquing writing, questions about uploaded documents, and creating documents beside the chat. Illustrative use: ask questions about a reading you are permitted to use, then have Claude turn your notes into an editable one-page study guide. Stanford provides a managed version whose features include a Learning Mode (see Guided learning below). (Stanford UIT: Claude for Education (opens in a new tab))
Gemini: Google’s assistant at gemini.google.com, used with your Stanford Google account, for explaining, drafting, answers drawing on Google Search, and longer Deep Research (opens in a new tab) reports. Illustrative use: ask for a first map of public writing on a topic, then open the sources it lists. Stanford provides the standalone app; its page says Gemini features built into Docs, Sheets, Slides, and Gmail are currently disabled for Stanford accounts. (Stanford UIT: Google Gemini (opens in a new tab))
Stanford AI Playground: A Stanford-managed platform for trying models from several providers in one place, with your SUNet ID. Illustrative use: send the same question about a public reading to two models and compare how their answers differ. It is a separate service: using a Gemini or Claude model there is not the same as using Google’s or Anthropic’s own app, so check the Playground for the features it offers. (Stanford UIT: AI Playground (opens in a new tab); see also the Hub’s Playground guide.)

AI also appears inside everyday software, but what you get depends on the license. Stanford’s free Microsoft Copilot Chat, for instance, is a separate chat tool; the paid Microsoft 365 Copilot add-on is what places AI inside Word, Excel, and PowerPoint. (Stanford UIT: Microsoft Copilot Chat (opens in a new tab))

Four separate questions apply to every feature below: does the provider offer it; does Stanford provide it; is it on in your account; and is your use, with your information, permitted? As Google puts the third point for school accounts, “feature availability and how your data is handled depends on your Workspace license.” (Google, “Use Gemini Apps with a work or school Google Account” (opens in a new tab)) Which services are approved for which kinds of information is on Stanford’s Approved AI Services (opens in a new tab) list.

Feature guide

OpenAI’s ChatGPT capabilities overview (opens in a new tab) is a useful single-page tour of one product’s features; other providers document similar features under their own names.

Ask and explore

Conversation and follow-up: What it is: asking, then refining with follow-up questions in the same chat.
SLS example: “Explain the difference between these two terms in this public reading,” then “Give me a new example.”
Not guaranteed: that the explanation is correct; fluent answers still need checking.
Model choice and thinking settings: What it is: some products let you pick a model, or a setting that spends more computation working through a problem before answering.
SLS example: choosing a slower setting for a multi-step logic question.
Not guaranteed: new information. More computation is not new source access.
Guided learning: What it is: modes that coach rather than answer, for example Claude’s Learning Mode as described by Stanford.
SLS example: working through a practice problem with hints instead of a model answer.
Not guaranteed: that the hints are right, or that the use is allowed in your course.

Bring information in

File upload: What it is: giving the assistant a document to read, such as a PDF or slide deck.
SLS example: asking questions about a public report.
Not guaranteed: that every part of the file was read (Lesson 4).
Web search: What it is: the assistant looks up current web pages and can cite them.
SLS example: finding the official page for a public event.
Not guaranteed: that the cited page says what the answer says.
Research modes: What it is: longer, multi-step searching and synthesis, called deep research, Research, or Deep Research depending on the product (Lesson 3).
SLS example: a first map of public writing on a topic.
Not guaranteed: completeness, or access to subscription databases.
Connected sources: What it is: connections that let an assistant read material in another service, such as a cloud drive. Products call them connectors, apps, or plugins. Making a connection available is one step; authorizing it to reach your account is another. (OpenAI: Connected apps in ChatGPT (opens in a new tab)) Stanford reviews connectors before allowing them; see its connectors page (opens in a new tab), which also asks you to check with your department first.
SLS example (hypothetical): finding this week’s meeting agenda in a shared folder your office has authorized for this use. This course never asks you to connect an account.
Not guaranteed: that a connector being available means your use of it is appropriate, or that it can see everything in the other service.

Create and revise

Drafting and critique: What it is: producing a first draft, or commenting on yours.
SLS example: tightening a public event announcement.
Not guaranteed: accuracy, or that the use is permitted for coursework.
Side-by-side editing: What it is: a pane beside the chat where a document or code is edited, called Canvas in ChatGPT and Gemini and artifacts in Claude. (Anthropic: What are artifacts? (opens in a new tab))
SLS example: revising a one-page study guide paragraph by paragraph.
Not guaranteed: the same behavior in every product; Canvas and artifacts are related, not equivalent. Creating content, retaining it in a service, exporting a copy, and sharing it are different events, and some services save work automatically (Anthropic says artifacts are saved to an Artifacts tab and start private to you). Check where the result is stored, who can access it, and whether you have exported the version you need.
Files and data analysis: What it is: creating documents, slides, or spreadsheets, and in some products running code to analyze data.
SLS example: turning a small public table into a chart.
Not guaranteed: a correct figure. A displayed calculation is not proof that one was run, and one that did run can still use the wrong range, formula, or units. OpenAI advises reviewing the code, outputs, and assumptions before relying on a result; for a small table, recompute a figure yourself. (OpenAI: Data analysis with ChatGPT (opens in a new tab))
Images: understanding versus creating: What it is: describing an uploaded image, or generating a new one; two different features.
SLS example: describing a public chart; drafting an event graphic.
Not guaranteed: that a description caught small text, or that you have rights for a generated image’s use.
Voice: What it is: speaking and listening instead of typing.
SLS example: rehearsing a short oral explanation.
Not guaranteed: anything about who is on the other end. Voice is an interface, not a person.

Organize and personalize

History: What it is: your past conversations, which many services save to your account so you can return to them; OpenAI, for example, says regular chats remain saved until you delete them. (OpenAI: Chat and file retention (opens in a new tab))
Not guaranteed: that an old chat informs a new one.
Memory: What it is: information the assistant carries across conversations to personalize later answers. Whether it is available, how it works, and what you can view or delete depend on the product, your plan or workspace, and your settings.
Not guaranteed: that it remembers what you expect, or forgets what you want forgotten.
Projects: What it is: a container within an application that groups related chats, files, and instructions for one ongoing piece of work. A project is not the same as a workspace, the organization’s managed environment (Lesson 1).
SLS example: a semester reading group with the syllabus and standing instructions.
Not guaranteed: that each chat can see the others. Anthropic, for example, says a project’s knowledge is used across its chats and describes a separate project memory where memory is enabled. Put important material where the product expects it, such as the project’s files. (Anthropic: Projects (opens in a new tab))
Saved instructions, custom assistants, skills: What it is: related but different reusable setups. Saved instructions apply to your chats generally. A custom assistant, such as a custom GPT or a Gemini Gem, is a configured assistant with its own instructions and sometimes files. A skill, such as a Claude Skill, packages a procedure the assistant can apply to a task.
Not guaranteed: that someone else’s setup is accurate, suits your task, or has been checked. The Hub’s AI Skills are examples built for SLS.

Act or repeat

Some assistants can take actions: use connected apps, carry out multi-step tasks, or run scheduled tasks. Keep one distinction in view: an assistant that writes a proposed email has not sent it, and a chat that says “I’ll remind you” has not necessarily created a reminder. Actions like these depend on permissions and are covered in the Advanced level. None is needed for this course.

Try it: match tasks to features

Constructed, low-risk tasks. More than one product could fit each.

A staff member needs a two-paragraph summary of what a public university web page says about an upcoming holiday closure, with the page cited.
A student wants to revise a one-page study guide on public course material, and keep an editable version to change later.
A faculty member runs a reading group all semester and wants the syllabus and standing instructions available each time they ask for discussion questions.
A staff member has 40 names in a list and needs them in alphabetical order.

For each, which feature or features matter most?

Answers and discussion

A source: web search that cites the page, or pasting the page text into the chat. Either way, check the summary against the page itself.
An editable output: a side-by-side editing pane (Canvas or artifacts) or document creation. Check where the result is stored, and export an editable copy, such as a Word document, if you will revise it outside the tool.
Ongoing context: a project holding the syllabus and instructions. Memory is a different feature, and whether it applies depends on the product and settings; a project puts material tied to one piece of work where the product expects it.
Ordinary software is enough. A spreadsheet or word processor can sort the list without generative AI. Choose the intended sort key and range, and check that every entry remains present, related fields stay together, and the resulting order is the one you intended.

Notice that none of the answers names a “best” assistant. Several products offer each feature; which one you use depends on what Stanford provides, what your account has switched on, and what your task and information allow.

A feature name describes a design, not a result. Search does not guarantee the page supports the answer; a project does not guarantee memory; creating a file does not tell you where it is stored, whether a copy has been exported, or who can see it. Try features with public material first, and check the output.

Resources

OpenAI: ChatGPT capabilities overview (opens in a new tab) — provider documentation; one product’s features, which vary by plan and settings.
OpenAI: Data analysis (opens in a new tab), chat and file retention (opens in a new tab), and connected apps (opens in a new tab) — provider documentation on reviewing analysis, how chats and files are kept, and app permissions.
Anthropic: artifacts (opens in a new tab) and projects (opens in a new tab) — provider documentation.
Google: Gemini Apps Help (opens in a new tab) — provider documentation.
Stanford UIT: Approved AI Services (opens in a new tab) and Available Connectors (opens in a new tab) — what Stanford provides, and for which information.
Self-Paced Courses — the Hub’s own courses and tutorials, and Learning Beyond SLS for providers’ own training.