New to AI tools, or just want the jargon decoded? This is every term you will meet in Smith and across AI generally, explained in plain language. Skim it once, or keep it open in a tab and look things up as you go. Terms are grouped by what they relate to.
AI Basics
The general vocabulary of AI tools, useful anywhere, not just in Smith.
An AI that takes a series of steps toward a goal instead of giving a single answer.
The engine that reads your prompt and writes a response. Smith offers several, and the ones you can pick from are the models your project has activated. The line-up changes as new releases land, so the Settings panel is the current list.
How much a model can hold in mind at once: your prompt, your selected files, and the conversation so far.
A way of turning text into numbers so the AI can find which of your documents are most relevant to a question. The quiet engine behind file search.
Giving the AI one or two examples of what you want, so it matches the pattern. Showing beats describing.
Training a base model further on specialized data to make it better at a narrow task. Done by model makers, not in everyday use.
Pointing the AI at your own files so answers come from your information, not just general knowledge.
A confident answer that is wrong. Grounding in your files and comparing models both reduce it.
The act of the model generating a response. When you hit send and the answer streams back, that is inference.
The date a model's general training stopped. For anything newer, ground it in your own current files.
Large language model, the technical name for an AI model. Same thing.
An AI that can work with more than just text, such as reading an image or a chart alongside your words.
What you type into the box: your question or instruction. Clearer prompts get noticeably better answers.
The craft of writing prompts that get good results: being specific, giving context, and saying what format you want back.
Retrieval-augmented generation. A method where the AI looks things up in your documents before answering, instead of relying only on what it already knows. It is what makes grounding work.
A model built to work through problems step by step before answering, stronger on complex analysis, usually a little slower.
Background instructions that shape how the AI behaves across a whole conversation, set before you start typing.
The setting for how safe or adventurous responses are. Low is precise and predictable; high is bold and varied.
The small chunks of text a model reads and writes, roughly three quarters of a word each. This is why very long documents cost and use more.
The large body of text a model learned from before you ever used it. Your Smith uploads are never part of it.
Smith Features and Concepts
The parts of Smith you will see and use day to day.
Models beyond the included set. An advanced model is purchased and then activated for your project before anyone can use it, and it shows as locked until that happens. Newer GPT and Claude generations, including Claude Opus, Sonnet and Fable models, arrive here, so the set grows over time and the Settings panel shows what your project has right now. Models carry cost indicators, so you can see the relative cost before you pick one.
A way to tailor Smith's output for a specific group you write to often. Coming soon.
The sources Smith attaches to a grounded answer, so you can trace each claim back to your own documents.
The base models that come with your project and are displayed as included. Anything outside that set is an advanced model, which has to be purchased and activated before it can be used.
A saved profile of your role, work, and goals that Smith references, so answers come back with your context already built in.
Smith's ability to carry useful context within your work, so you repeat yourself less. Managed from the settings gear.
The dropdown below the prompt box where you pick which model handles your request. It lists the models your project has activated, and you can change it anytime.
Running the same prompt across two or three models and comparing the answers. It is off by default, so switch it on when you want the comparison. One answer is a guess; three is a cross-check.
Your own area in Smith, shown in the left panel. Where you keep and select the documents Smith can draw on, and where you manage your saved prompts and folder access.
The button in the persona builder that fills in every field based on the few you entered. The fastest way to a complete persona.
Saved instructions that shape how Smith responds: its role, tone, and rules. Build it once, use it in any chat.
The dropdown next to the model selector where you switch on a saved persona for the current chat. Your choice stays in place for the rest of that chat.
The underlying instruction a persona sends to Smith on your behalf. It is shown on the persona card, so you can see how a persona will behave before you switch it on.
Ready-made prompts in Smith you can run with one click, found in the tab next to Custom. You are not limited to the supplied ones: you can create your own prompts and manage them from My Workspace.
One chat where you jot down recurring tasks as you notice them. It becomes the raw material for automating your week.
The answer text that appears piece by piece right after you send a prompt, while the model is still working. Seeing it stream means Smith is responding in real time.
A short description of how you write, distilled by Smith from your own writing. It powers the My Voice persona.
One conversation with Smith. Follow-ups in the same thread keep all the context that came before.
Files you add to a conversation. Smith keeps these separate from the files in your workspace, and attachment selection shows the two groups apart.
A running measure of how much your project has used Smith. It is calculated month to date, so the figure you see covers the current month so far.
Your space in Smith where your files and folders live. Who can see or change a folder is set by its permissions. Nothing in it is used to train the models.
The files kept in your workspace folders. When you choose files to attach, these are listed separately from anything you have uploaded to the conversation.
Working Well with Smith
Practical techniques that make every session better.
Pasting a sample of the output you want, a past email, a report you liked, so Smith matches its shape and tone instead of guessing.
Using the output of one step as the input to the next, building a bigger result from a sequence of moves. The idea behind recipes.
Telling Smith the limits up front: length, format, audience, what to leave out. Tighter asks come back closer to ready.
Improving an answer by replying in the same thread rather than starting over. Smith keeps the context, so each round gets closer.
Telling Smith who to be for a task, such as a careful editor or a skeptical reviewer, to shape the kind of help you get.
Asking Smith to cite where each claim came from, then checking the citations, so you trust the answer before you use it.
Next Steps: Ready to put these into practice? The Getting Started guide shows the screen these terms describe, and the Playbook puts them to work.
Related Articles
Getting Started with Smith
See the interface and features these terms describe.
The Playbook
Put the concepts to work with single-move plays.
Learning Tracks
A guided path by skill level once the vocabulary clicks.
Comments
0 comments
Article is closed for comments.