A small group inside the Applied AI Club that builds software for other student organizations at Penn State.

Do I need to know how to code?

No, and most of us do not. We work with AI tools that write most of the code, and you learn by using them on something real.

What would I actually work on?

Whatever another club needs. Usually a website, a small tool, or something that broke and nobody has time to fix.

How much time does it take?

A few hours a week during the semester.

Any Penn State student with a psu.edu email address, undergraduate or graduate, in any major.

Does it cost anything?

No. Labs is free to join, and free for the student organizations we build for.

Applied AI Labs at Penn State

What is Labs?

What is Labs?: A small group inside the Applied AI Club that builds software for other student organizations at Penn State.
Do I need to know how to code?: No, and most of us do not. We work with AI tools that write most of the code, and you learn by using them on something real.
What would I actually work on?: Whatever another club needs. Usually a website, a small tool, or something that broke and nobody has time to fix.
How much time does it take?: A few hours a week during the semester.
Who can join?: Any Penn State student with a psu.edu email address, undergraduate or graduate, in any major.
Does it cost anything?: No. Labs is free to join, and free for the student organizations we build for.

AI models

Nine models from five companies. These are the ones we plan to work with and learn our way around. Which we actually use depends on how Labs funding is settled, so read this as the shape of it rather than a final list.

Anthropic

Claude Fable 5

The most capable model here and the most expensive by a wide margin. Worth reaching for only when a problem is genuinely hard.

: $10
: $50
: 1M tokens

Prices are US dollars per million . Hover any underlined term for what it means.

Words worth knowing

Token: The unit models count text in. A token averages about three quarters of a word, so a million tokens is roughly 750,000 words. Pricing and limits are all expressed in tokens.
Context window: How much a model can hold in mind at one time, counted in tokens. Anything past the limit it cannot see, so a long document sometimes has to be broken up. Bigger is not automatically better, but running out is a real problem.
Input and output: Input is the text you send a model, output is what it writes back, and they are priced separately. Output usually costs several times more, so a long answer is far more expensive than a long question.
Open weights: Weights are the file holding everything a model learned in training. Most companies keep theirs private, so the only way to use the model is to pay per use. A few publish theirs, which means anyone can download the model and run it themselves.
API: A way for one piece of software to talk to another with no person in between. When a website answers questions using a model, it is calling an API rather than typing into a chat box.
Agent: A model given tools and permission to use them, so it can work through a whole task on its own instead of answering one question at a time. This is most of what Labs builds with.

AI harnesses

The program you work with a model through is called a . Labs members use Hermes Agent, from Nous Research, each running their own copy pointed at the club’s account.

Memory

Persistent memory

It keeps what it learns about a project between sessions, so you are not re-explaining the same codebase every time you sit down. Once it has solved a problem it does not forget how.

Survives restarts
Learns the project
Remembers past solutions
Stored on your machine

Hermes Agent is open source under the MIT licence. Everyone runs their own copy, so nobody is waiting on a shared seat, and the club pays for the model behind it rather than for the software.

Working with a harness

Harness: The program you work with a model through. The model itself only reads and writes text; the harness is what gives it a terminal, files, and tools, and decides what it is allowed to touch.
Tool call: When a model asks the harness to actually do something, like run a command or read a file, rather than just describing it. Everything an agent accomplishes it accomplishes through tool calls.
Skill: A procedure the agent saves after working something out, so it can repeat it later without solving it again. Skills are why the second attempt at a task costs less than the first.
Subagent: A second agent the main one starts to handle part of a job, with its own conversation and its own terminal. It reports back a result rather than everything it read to get there.
Sandbox: An isolated place for code to run where it cannot reach the rest of your machine. The safety of letting an agent run code it wrote is almost entirely a question of whether it is in one.
Model agnostic: Not tied to one company’s models. A model agnostic harness lets you change which model is behind it without changing how you work, which matters when prices and rankings move every few months.

Apply

Everything above is what you would spend the semester learning to use. You are not expected to know any of it yet, and most of us do not.

We are taking 9 people this fall. Applications close September 4 and the first teams start the week after. Plan on a few hours a week from then until the end of the semester.