Parlor Study

Turn reading and lecture notes into ideas you can explain in your own words. AI-assisted discussion and practice. Web preview · Request an invitation ↓

[Parlor Study @parlorstudy]

Parlor Study asks you to explain what you read in your own words, then tells you which part of your explanation is off. Lecture notes, concept maps, and book discussions for the classics. Web preview by invitation: https://t.co/dEC5vAJDMo

[Parlor Study @parlorstudy]

@AlexJQuigley @cbokhove 'Check for understanding' is a good test case. The hard part to adapt is what counts as evidence in each subject: in chemistry a correct definition can hide a weak model, so the check has to ask for a prediction or a why. That's subject knowledge, not a generic technique.

[Parlor Study @parlorstudy]

Nearly a thousand high-school math students: those given a basic GPT-4 tutor did better in practice, then got 17% lower grades on exams without it. A tutor with teacher-written safeguards largely avoided that.

What the study does and doesn't show: https://t.co/Z30aYCYpy6

[Parlor Study @parlorstudy]

@AlexJQuigley The 'can't predict the return' point feels central. Retrieval costs effort now and pays off weeks later, so students often rate rereading as better even after retrieval worked better for them. Seeing the gap on their own delayed quiz may persuade more than being told.

[Parlor Study @parlorstudy]

Getting a flashcard right once feels like the moment to drop it.

In Karpicke and Roediger's 2008 study, students who kept testing themselves on words they'd already recalled remembered more a week later. Restudying them didn't help the same way.

https://t.co/pqDGQWNxaZ

[Parlor Study @parlorstudy]

@daisychristo The detail I didn't expect: the biggest disagreements were mostly human slips, not AI ones, so the AI ends up working as a check on tired markers. The handwriting step sounds like the part to watch, since a bad transcription changes what's being judged.

[Parlor Study @parlorstudy]

The Great Gatsby takes about 3 hours to read. Frankenstein, just under 5. Pride and Prejudice, under 8. Middlemarch, a little over 20.

Reading times for 100 classics, counted word by word, with a time for every chapter and a daily plan:

https://t.co/rzqHUFPJ5L

[Parlor Study @parlorstudy]

@AlexJQuigley The order is the whole thing: Leah writes her own answer from memory first, then checks it against the tool. Same AI, opposite effect. The catch, as you say, is that cognitive misers won't pick that order on their own, so it has to be built into the task.

[Parlor Study @parlorstudy]

Instructors: what's one idea students can define on a test but can't explain when you ask them to? For intro stats it's usually the confidence interval.

[Parlor Study @parlorstudy]

@EnserMark The idea that a principle has to rule something out is what I'll take from the whole series. "Support is temporary" rules out the Year 11 frame. "Support pupils to write well" rules out nothing. Thanks for doing it in the open, sources and all.

[Parlor Study @parlorstudy]

@EnserMark The three flood sentences make the point well: more precise, not longer. Step 2 is the test I'd want everywhere. A short piece with no frame and no starters shows quickly whether the frame was supporting the writing or doing it.

[Parlor Study @parlorstudy]

Victor Frankenstein spends two years building a "beautiful" creature, then runs the moment it opens its eyes.

Our free Frankenstein study guide has discussion questions for every stage of the book.

https://t.co/daFPdAcmkP

[Parlor Study @parlorstudy]

@C_Hendrick What happens after "help me understand this" is the part to watch. An AI explanation reads clearly, so it feels understood. You only find out whether it was once the chat is closed and the student tries to explain it back. Instruction could build that step in.

[Parlor Study @parlorstudy]

@EnserMark The pupil who said she hadn't really improved, she'd just done what she was taught, is the line that stuck. Learning the method gets counted as cheating, not as getting better. Naming what she did that made it work (Step 4) seems the most direct answer to that.

[Parlor Study @parlorstudy]

Why does highlighting feel so productive, and why does it help so little?

We wrote up what the research says, footnotes included, and how it shaped Parlor.

https://t.co/ks1p25pKvG

[Parlor Study @parlorstudy]

@EnserMark The bit about "and" is what I'll keep: it says two things happened, not that one caused the other. The waterfall answer has the knowledge and none of the causation. Splitting a pupil's own run-on into lines and asking which causes which makes that easy to see.

[Parlor Study @parlorstudy]

@EnserMark The second waterfall conversation is the one that fools everyone, the pupil included. Useful that you flag the check as an untested rule of thumb; it makes the rest easier to trust.

[Parlor Study @parlorstudy]

891 students' take-home exams: those who wrote their own answer first and then asked AI for feedback got the highest grades. Those who handed the work to AI did worse than students who didn't use AI at all.

https://t.co/YXdTIDi6YA

[Parlor Study @parlorstudy]

@EnserMark The point about conversation being cued stuck with me. "So what happens at the bend?" already does half the retrieval. Looking forward to the "say it out loud with no help" test in part two.

[Learning & the Brain @learningandtheb]

You are looking for an awesome podcast on #Checking_for_understanding.

Look no further -- @dylanwiliam, @teacherhead, @nsachdeva2019, &
@KateJones_teach are here to help:

https://t.co/NVuE9WAqJM