Bootcamp / Session 1 / 1.1
Session 1.1 · Opening · Alper Kucukural
Welcome. Here is the question we will keep asking for six weeks.
When AI gives you an answer, how do you know whether to trust it, and what do you do about it? Today three people answer that question from three angles.
Recording
Why this bootcamp exists
This is the successor to the Sequence Analysis Bootcamp. The old course spent its first weeks teaching you to produce code. That part got cheap.
“Producing the code is cheap. Understanding the code is more expensive now.”
Eric Ma
Today in three acts
Think of AI as an aircraft. First you learn how the plane works, then how an experienced pilot flies it, then the flight rules and the airport you will fly from. By the end of the afternoon you can explain why a language model is sometimes confidently wrong, and you know the rules and habits that keep that from reaching your results. Times are approximate.
The question for today and how the afternoon fits together.
Next-token prediction, temperature, training, hallucination, foundation models.
Part 1.2 →Setup help is available if any of your green lights is not working yet.
AI coding agents, a real analysis workflow, reproducibility, a live Claude Code demo.
Part 1.3 →The rules for our data, guardrails, how I use this every day, and what comes next.
Part 1.4 →Keep this checklist open all afternoon
Nine questions to ask before any AI output enters your work. Each speaker today answers some of them. At the end (1.4) we fill in the rest, and it becomes your homework.
- Could this be a fluent guess rather than a fact?Answered by Eric (1.2)
- Would I get the same answer if I asked again?Eric (1.2) and Tommy (1.3)
- Could this be out of date?Eric (1.2)
- Did I check it against the primary source or the raw data?Eric (1.2) and Tommy (1.3)
- Do I know enough to notice if it is wrong?Tommy (1.3)
- Is there anything here that should not leave our walls?Alper (1.4)
- What was the AI allowed to do, and what did it do?Tommy (1.3) and Alper (1.4)
- Could someone else trace how this result was made?Tommy (1.3) and Alper (1.4)
- Would I be comfortable disclosing exactly how I used AI here?Alper (1.4)
Ground rules and what to expect
In the room
- No patient or unpublished data in any AI tool today. Everything we show uses public data or our own test runs.
- Every session is recorded. Materials and links live on these pages and in the GitHub repository.
- Laptops open, but you do not need to type along. The live demos are for watching. Hands-on work starts with the domain sessions.
- Questions are welcome. Ask during or at the end of each part.
Before next Friday
Setup is pre-work. Get your four green lights (Foundry Connect login, SSH to the cluster, Foundry Connect running jobs on the cluster, and Claude connected to Foundry Connect) before Session 2. The steps take about 30 minutes, plus the wait for your HPC account approval.