## Welcome to Seminar 1 **Data-Driven Life Sciences 2026** * Good morning — and welcome to our first live seminar. * You made it through the first computer lab yesterday. That was a real lift. * This morning: what the lab was for, **what you told us**, and how these seminars work. * We're figuring this out together — bring your questions. ----- ## Agenda * **Part 1 — The computer lab:** what it trained, what the goal really was, and a full debrief of **your feedback**. * **Part 2 — The seminar:** what it's for, how presentations work, and how to prepare. * Plenty of room for **questions** along the way. ----- ## What this course actually trains * We are training the **forward-deployed scientist**. * You take **someone else's** biological problem. * You **interview** them until the question is precise. * You **direct an AI agent** to do the labour — while **you** supply the judgement. * The loop: **specify → direct → verify.** * The scarce skill is **specification, direction and verification** — *the harness, not the model.* ----- ## The two-agent lab (recap) * **Agent A — the data owner.** You **interview** it to pin down a precise question. It answers a good question honestly and a vague one uselessly. * **Agent B — Pi, the analyst.** You **direct** it via `AGENTS.md` / spec to do the actual work. * **You are the only channel between them.** Nothing passes from A to B except what you write down. * That channel **is** the examined skill. ----- ## What the goal of Lab 1 REALLY was * To go through the **whole loop once** and get comfortable with the machinery: * **Interview** Agent A → **use the portal** → **set up Pi** locally → **direct** it → produce **something** end to end. * That's it. One full pass, hands on every part. * The aim was **familiarity with the pipeline** — not the result at the end of it. ----- ## What the goal was NOT * **NOT** a deep, polished, or publication-grade analysis. * **NOT** finishing with time to spare. * **NOT** completing every validation step. * A **rough, rushed, incomplete-feeling** first pass is completely fine — and expected. * **Week-1 quality is not graded.** ----- ## Lab 1 — by the numbers * **20 of you handed in a full lab** — the whole loop, interview to delivered report. * **16 seminar decks** prepared for today. * **16 of you left detailed feedback ratings** — thank you, we read every word. * How you rated it (out of 5): * **Portal — 4.3** · **Design — 4.2** · **Instructions — 4.0** * **Overall — 3.6** * A hard, deliberately-stretchy first lab landing around **3.6 overall** is a **healthy, honest** signal — not a disappointing one. ----- ## What you did remarkably well * You **refused vague words.** You made Agent A define "trustworthy", "unusable", "scaffold class" instead of guessing. * You asked the **killer question** — *"what would make you NOT trust this result?"* — and turned the answer into your must-nots. * You **read the brief back** to Agent A to check your own understanding. * You caught **data traps**: CSV headers that didn't match the description, leakage (island / sample-id → species), temporal splits, "blanks are never zero". * You wrote honest **"what I would NOT claim"** sections. * You **directed** Pi rather than trusting it — pushing back on odd rankings, auditing its own spec. ----- ## What you told us — the honest feedback We grouped everything you wrote. Five themes came back again and again: * **Setup ate the clock** — installing Pi, the terminal, Windows especially. * **Time vs. careful checking** — "I had to rush the validation steps." * **A confusing start** — "at first I wasn't sure what to even ask." * **It felt stressful** — "a lot of us didn't finish in time." * **Online, it's hard to know** you're on the right track. Let's take them one at a time. ----- ## "Setup ate the clock" * By far the most common note — Pi, the terminal, permissions, **Windows** in particular. * You're right, and it's a **one-time cost**: Pi and the portal are set up **now**. Next week you walk in already installed. * **What we're doing:** an OS-by-OS **setup checklist** (incl. Windows) so the lab window goes to *thinking*, not *installing*. * The skill we're grading was never "can you install a CLI tool." ----- ## "Not enough time to check carefully" * Several of you felt the tension between the clock and doing verification properly. That tension is **real**. * But remember: **Week-1 quality is not graded.** Rushing the last validation step in week 1 is **expected**, not a failure. * As setup time disappears, **more of the window goes to the actual loop** — interview, direct, verify. * Noticing *"I didn't have time to check this"* and **writing it down** is itself the skill. Many of you did exactly that. ----- ## "I wasn't sure what to even ask" * A confusing start — *"what do I ask Agent A?"* — came up a lot. * Here's the thing: **that confusion IS the job.** Sitting with someone else's problem and not yet knowing the right question is what a forward-deployed scientist does every day. * It is **the hardest part**, and it is **the point**. It gets easier fast. * **What we're doing:** tightening the lab's opening instructions so the *first move* is clearer — without removing the productive struggle. ----- ## If it felt too long — that's okay * Feeling rushed was **never the bar**. The point was the **process**, not the polish. * Struggling to ask Agent A the **right question** at first? That is exactly the skill we're practising — **not a failing**. * You are supposed to be a bit uncomfortable in week 1. **It gets easier.** ----- ## A bit of stress — on purpose * We hear that this is **more intense** than a slow, week-long lab. That is a **deliberate design choice**, not an oversight. * We believe **dedicated, focused time — plus a little pressure — is how this skill actually sticks.** Comfortable and diffuse doesn't build it. * But we also **care that it's manageable**: you may **miss one** mandatory session (a lab *or* a seminar). Just email **ddls-course@scilifelab.se** *before* the session. * If it tips from stretching into overwhelming, **tell us** — that's data too. ----- ## Why the lab comes out the same week * A few of you asked us to release the material a week early. We thought about it — and we keep to **same-week release on purpose.** * Each week's lab is shaped by the **previous week's feedback** — so *when* we release is itself a **data-driven decision**. We'd rather not release it blindly. * Some weeks bring a **live guest lecture** — handing out the lab before the topic is introduced wouldn't make sense. * The lab runs on the **live portal** — we can watch it and keep it healthy in the **fixed lab window**, not around the clock for a week. * And yes, it's **more intense** than having days to prepare. That part is deliberate: **focused time + a bit of pressure is how this skill sticks.** ----- ## "Online, it's hard to know if I'm on track" * Real, and worth naming: no one is leaning over your shoulder. * That's exactly what the **live lab window**, this **seminar**, the **portal coach**, and us **watching the portal** are for. * **Ask early and ask often** — in the lab window, in chat, here. If you hit it, others did too. ----- ## Small things you asked for You gave us concrete portal ideas. A few we're taking away: * A **"explain this simply"** helper / glossary for jargon in Agent A's replies. * A **stopping point** for the suggestion "wand" so it doesn't loop forever. * Clearer signals that you're **on the right track** during the lab. Keep them coming — the portal is **ours to shape together**. ----- ## Where the labs stand — your feedback * **We read every submission.** All of them. * We're sending each of you **individualised written feedback** on your lab: * what you did well, * what to improve, * what to be aware of. * It's **AI-drafted** from your actual interview + files, then **reviewed by the teaching team** before it reaches you. * Watch your **portal bell / dashboard**. ----- ## Questions about the lab? * Let's talk. * What got stuck? What was confusing? What surprised you? * No question is too basic — if you hit it, others did too. ----- ## Part 2 — The seminar **How this Friday session works** ----- ## What the seminar is for * First, to **review the computer lab together** — a shared debrief (that's what we just did). * We look at what worked, what tripped people up, and what to carry into next week. * **Then** we move on to selected presentations. * Think of it as a weekly checkpoint, not an exam. ----- ## How presentations work * Each week, **~7–10 students are drawn at random** to present. * You present the work **you** did in **that week's** computer lab. * **7 minutes + 3 minutes** of discussion. * This **replaced** the old "journal club" — you present your own work, not a paper. ----- ## The core idea ## *"You can outsource your thinking, but not your understanding."* ----- ## Outsource the doing, not the knowing * We made it **easy to produce** things — the AI writes the report, even makes the slides. * That does **not** make it any easier to **understand** what was done. * To present it **orally**, you have to actually understand it. * So preparing a talk starts from **understanding your own work** — not from the slides the AI made. ----- ## What this means for you * **Before** the seminar, spend time understanding **your own report**. * Be ready to explain: * your **question**, * your **data**, * what the **analysis** actually did, * what you **would** and **would not** trust. * **To pass the seminar you must be able to present your work if you are selected.** ----- ## Let's begin * You did the hard first pass. Well done. * Today: debrief together, then a few of you present. * Bring your questions — this room is for figuring it out. * Let's get started.