A study published this year gave 1,339 teachers the same piece of student work with a wrong grade already attached. When the grade was too harsh, teachers left it standing more often if the label said it came from AI.

It made me think about how often a number reaches us before our own judgment does.

Think of the last time someone else's grade was in front of you before you read the work, from a TA, a colleague or a tool. Did you change it, or did it stand? One line is plenty.

Midterms are going back this week. Before you return yours, try sorting the stack by question instead of by student.

When a lot of students give the same wrong answer, it is not random. It tells you which step broke, and that is the step to rebuild before the class moves on.

Today's Teaching Day session turns that into one page you can post the same day the grades go back: a short glossary, one worked example with the reason beside each step, and three practice problems.

Question for today, and one line is plenty: what was the most common wrong answer on your midterm this term?

Today's question, answerable in one sentence.

Think about the next writing assignment you will give. There are four kinds of AI help a student could ask for: planning, drafting, reviewing, or the whole draft.

So: which one would you close first on that assignment?

One word is plenty. A one line reason is a bonus.

Today's question, answerable in one sentence.

It is midterm week for a lot of you. Rereading feels like studying because the page looks familiar, but familiar is not the same as remembered. The fastest way to find out what you actually know is to close the book and explain it out loud.

What is one concept you were sure you knew until you tried to explain it out loud?

One line is plenty. If you are teaching this week, the same question works on your own course material.

Today's question, answerable in one sentence.

Alaska's draft AI policy says AI can recommend, but a human decides. A checker score, a drafted comment and a suggested grade all count as AI recommending.

So: name one place in your course where software recommends something about a student right now.

One line is plenty.

Today's question, answerable in one sentence.

Midterm stacks are in. A few of those essays will not sound like the student who has been sitting in row two since August, and a detector score will not settle it. A score is a likelihood about text, not a finding about a person.

What holds up is evidence from the student's own words, held against your own rubric. Four lines: the rubric line, the quote, the gap between them, and the next step for the student.

One answer to start the thread: reread the rubric line for that criterion before you reread the paper, so you test the work against what you actually asked for.

So, one line is plenty: when a paper worries you, what is the first thing you do before you say a word to the student?

Today's question, answerable in one sentence.

Dartmouth just told its faculty that a detector score can never carry an AI case by itself. The committee will look at evidence beyond the score.

So, for your next major assignment: if a detector flagged one paper tomorrow, what is the first piece of evidence you would want besides the score?

A draft, a source log, a conversation, something else? One line is plenty.

A Berkeley working paper by Igor Chirikov found that after ChatGPT was released, the share of A grades in AI exposed courses rose by 13 percentage points. The increase was larger where homework carried greater weight.

Read that second sentence again. The inflation is worst exactly where the take home work counts most.

Yesterday Andreessen Horowitz launched a school with no homework at all, and an admissions process that asks for a portfolio instead of an essay.

So here is today's question, and it takes one sentence to answer.

What is the one assignment in your course you would be least surprised to see a model produce?

Not the one you are proudest of. The one that worries you. Name it and I will tell you the smallest change that puts the evidence back in it.

The students who stopped posting did not lose interest. They read the thread, saw the reply you gave someone else, and decided they had nothing worth adding.

You cannot prove that about any one student. You can check it tonight. Today's Teaching Day is about the reply, not the prompt: use their words, give one sentence of encouragement that is not a verdict, and ask a real question back that only they can answer.

The free session that walks you through three of them: https://www.harvestkernel.com/revive-the-thread/

One question for this space, and one word is a complete answer:

What is the first word of the last reply you posted on a discussion board?

Monday question, and it takes one sentence.

This morning's post is about the brief unsealed last week in the case against OpenAI and Microsoft, where Microsoft's CEO agreed under oath that chatbots give people the information instead of sending them to the source. For anyone who assigns readings, that is a design problem, not a discipline problem.

The fix I keep coming back to is small: build one class session around the passage a summary drops. The qualification, the example, the turn in the argument, or the author's voice.

So, one sentence: which reading this term would you least want your students to meet only as a summary?