---
name: ai-in-research-week-1-homework
description: Week 1 homework for UCLA's AI in Research class (EPSS 254, Fall 2026). Design the hardest task you can in your own field and package it as a Harbor / Terminal-Bench-Science benchmark task. Use when a student asks for help with this homework.
---

# Week 1 homework: the hardest task in your field

You are helping someone in UCLA's "AI in Research" class (a student, auditor or professor) with the week 1 homework, due **before class on Wednesday, October 7, 2026**. They will design **the hardest task they can in their own field**: one a top expert could do, and today's best AI agents likely can't. They will package it as a benchmark task that an AI agent attempts on its own, in a sealed container, graded automatically. You do the engineering; they bring the science. Along the way they learn how AI benchmarks work. Week 1 ends when their task is submitted as a pull request; the instructor then runs frontier agents on every task, so don't run one yourself.

Everything is in the class repository, **https://github.com/huangzesen/ai-in-research-homework**, including the full guide for you. Do this:

1. Say hello, and tell the student in two sentences what the homework is. **Before running anything, tell them what will happen**, so the person in front of the screen knows what's coming:
   - **Set up (once):** you check git and GitHub and install what's missing. They do one thing themselves, in their own terminal: log in to GitHub.
   - **Choose the task:** you ask about their research, and together you pick the hardest problem in their field that a program can still check.
   - **Build it:** you do the engineering; they write the instruction and decide what counts as a right answer. Nothing runs the task on their computer, so no Docker is needed.
   - **Submit:** they check what will become public, and you open their pull request. That's the end of week 1. The instructor runs everything from there: GitHub checks each task, then frontier agents try it.

   They sign off three times by typing "yes" in their own terminal. It usually takes a few sessions of 1–2 hours.
2. Find their copy of the class repository. If you're already inside a clone of `ai-in-research-homework`, or `~/ai-in-research-homework` exists, use it. Otherwise make one in the home folder: `cd ~` and then `git clone https://github.com/huangzesen/ai-in-research-homework.git`. Cloning needs no GitHub login; the class tools have the student log in right after. If `git` is missing: on macOS, running `git --version` offers to install it (the student clicks Install); on Windows, the student runs `winget install --id Git.Git -e` in PowerShell, then opens a new terminal; on Ubuntu, the student runs `sudo apt-get install -y git`. On Windows, work in PowerShell or Windows Terminal.
3. Make sure `uv` works (`uv --version`); it runs the class's tools. If it's missing, install it: on macOS or Linux `curl -LsSf https://astral.sh/uv/install.sh | sh`, on Windows (PowerShell) `powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"`. Then open a new terminal, or use the full path the installer prints.
4. In the clone, read the assignment, `week-1/README.md`. Then open `.agents/skills/build-a-task/SKILL.md`, **read the whole file**, and follow it from the beginning. It explains your role and every step, and has you run `uv run tools/hw.py`: a checklist that always says what to do next.

If you can't clone yet, read the same guide at https://github.com/huangzesen/ai-in-research-homework/blob/main/.agents/skills/build-a-task/SKILL.md.
