The Friday Harvest, Issue 002: Half Your Syllabus Does Not Travel

Three companies shipped student AI this week. The one that matters is the one that read your syllabus.

On Tuesday OpenAI launched a teen experience for ChatGPT. On Wednesday Google announced a student hub inside Gemini and shipped five learning features in Search. Every headline framed this as a story about students and cheating. It is not.

Two of those products take a syllabus as an input. The Gemini hub pulls the test dates and deadlines out of one and writes them into a student’s Google Calendar. Search accepts a syllabus as a source document for a study notebook. That is a change in what the document is. It used to be something a student skimmed. It is now something a parser ingests, before the student reads a word of it.

Here is the part nobody has said out loud. Extraction is lossy, and it is lossy in one direction. It reliably finds anything with a date attached. It does not find the late policy you wrote as a paragraph, or the sentence explaining why essay two comes before the midterm, or the reason the reading load drops in week nine. So the student arrives in week two with a calendar of your course that is completely accurate and missing the argument of it, and neither of you can see what fell out.

Call it the extraction gap. It is not a cheating problem. Nobody in that story is cheating.

August 19. Google announced a dedicated student hub in the Gemini app: study notebooks, flashcards, practice quizzes, and the ability to add test dates and deadlines to Google Calendar from a syllabus.

Your dates are going into student calendars this month whether or not you put them there.

The Verge

August 19. Google Search shipped five learning features. Practice quizzes now cover the ACT, AP, ENEM, GRE, JEE, LSAT, MCAT, NEET and SAT, built with The Princeton Review, Careers360, PhysicsWallah and Akira Enem. Lens will tell a student where their mistake is. Notebooks arrive in AI Mode, and Search will build study documents from uploaded class slides and syllabi.

The review materials your students use are now generated, free, and outside your view.

Google

August 19. Eligible students in the United States get one year of Google AI Pro at no cost, including 5TB of storage and higher Gemini usage limits. Students outside the US are offered Google AI Plus instead.

Cost is no longer the reason a student does not have the tool.

The Verge

August 18. OpenAI launched ChatGPT for Teens, which adds responsible homework reminders that try to recognise when a teen is trying to shortcut an assignment and redirect them into study mode.

A product now holds an opinion about what shortcutting your assignment looks like.

The Verge

ChatGPT for Teens. For most of you, do not bother.

This is the launch that got the coverage, so it is worth being clear about who it is actually for. ChatGPT for Teens applies to accounts identified as belonging to someone between 13 and 17, either by self reported age or by OpenAI’s own estimate. If you teach undergraduates, most of your students are 18 or over and will never see it.

Can you get it? No, and that is the honest answer rather than a complaint. There is nothing to enable, nothing to request from your institution, and no console where you can check whether a given student is in it. You cannot turn it on, you cannot turn it off, and you cannot see it.

What does apply to you is the piece underneath it. Study mode is not age gated, and the responsible homework reminders that got the headlines are a nudge into study mode rather than a block. A student can leave it. So if you read the coverage and thought some guardrail had arrived that changes what you need to do about assignment design, it has not.

Where this genuinely lands is dual enrolment and early college. If you teach high school students on a college roster, this is real, it is on by default, and it may change what your students report about their own tool use. Everyone else can skip it.

The syllabus has changed interface twice in living memory, and only one of those changes got noticed.

On Gamma’s higher education session, the instructor Damian Derino described how students used to get one: waiting outside a professor’s door to collect what he called “just a textbased printout,” then deciding from it whether to take the class. That document was the entire interface. It was read once, by one reader, in one sitting.

The first change was the move to the learning management system, which nobody experienced as a change in kind. Same document, different door.

This week is the second one, and it is a change in kind. The syllabus now has two readers and only one of them can raise a hand. Which means the design question is no longer only what will a student understand from this. It is also what survives being parsed, and those are different questions with different answers.

Source: Gamma Power Hour | Higher Education

Fifteen minutes, one syllabus, two columns. Do it before your first class of the week.

Open the syllabus you are actually teaching from. Not last year’s. Go through it once, top to bottom, and write every item that carries a date, a week number or a deadline into a list. Copy them, do not summarise them. That list is what a student’s calendar will contain by Friday.

Now go through it a second time and write a different list: every sentence that carries a policy, a standard, a reason or a sequence argument and has no date attached to it. The late work rule written as a paragraph. What separates a good answer from a complete one. Why the units are in that order.

Compare the two lists. The second one is your extraction gap, and for most syllabi it is close to half the document.

Then do one thing about it. Pick the single item from the second list that would do the most damage if a student never met it all term. Find the dated item it belongs to. Move it there, as one sentence attached to the date. That is it. One reason, relocated, so it travels with something a parser will find.

If you want the two column pass as a working page rather than a legal pad, we built one this morning and it is free: The Syllabus Extraction Test. Nothing is saved and no account is needed.

Then look at what you just did, because the arithmetic is the point. That is one reason moved, in one syllabus. Your second list probably has a dozen more on it. Multiply by the number of sections you teach, and again by the fact that a syllabus is not written once, it is rewritten every August, and the reasons get separated from the dates again every time you rebuild it. The work is not hard. It is the same work, repeated, and it never stays done.

The tool built for that specific job is the Syllabus Generator. It drafts from your Course Record rather than a blank page, so the policy, the reason and the date arrive attached to each other instead of three pages apart, and it exports accessibility first in nineteen languages.

And do the first one by hand anyway. You will only know whether the tool got it right because you already know what your own gap looks like.

Monday: An AI Certificate That Starts by Asking If the Problem Needs AI, plus The Bottleneck Brief.

Tuesday: the faculty teaching session, and Build Your Course Record By Hand.

Wednesday: AI Grant Proposals: Not Worse, Just Average, plus Twenty Minutes Before You Prompt.

Thursday: the student session on semester collisions, and The Collision Map.

Friday: AI Study Tools Just Shipped Without You. Your Syllabus Is the Input.

Send this to whoever in your department is writing the AI policy this term.

Not because it tells them what the policy should say. Because it tells them the thing the policy is going to miss: that the tools arriving on campus this month are not reading your students. They are reading your documents. A policy written about student behaviour will not touch that, and by the time anyone notices, the calendars will already be built.