Nobody learns this alone. Rooms are never empty here: alongside whoever is live, Mira seats anonymised answers from everyone who has taken the chapter before, plus peers she plays herself.
The room is always full.
Intro to SQL
Counting things in groups
Intro to AI Engineering
Answering from your own documents
The L&D view: not a completion rate — the wrong ideas the archive is still holding, in their words.
Intro to SQL · When a value was never filled in
62%
“empty means zero, or means an empty bit of text”
“Blank is blank, surely? If nothing is there it didn't happen.”
Intro to SQL · Turning many rows into one per group
41%
“I can still ask for any column I like once I have grouped”
“The customer name is still in the table. Why can't I just select it?”
Intro to AI Engineering · Why it needs your documents handed to it
71%
“we need to train the model on our data”
“Training is how you teach it things. That's the word everyone uses.”
Join a live peer, an echo replayed from the archive, or a partner Mira plays from the same wrong idea.
Counting things in groups
AI-facilitatedTalk Jonas out of empty meaning zero
Jonas thinks “empty means zero, or means an empty bit of text”
Counting things in groups
Echo replayReplay: a filter that keeps the benefit of the doubt
a product manager thinks “a filter keeps a row unless it clearly fails the test”
Counting things in groups
LivePriya still wants every column after grouping
Priya thinks “I can still ask for any column I like once I have grouped”
Counting things in groups
AI-facilitatedTalk Jonas out of LIMIT making it faster
Jonas thinks “keeping only five makes the query faster because it stops early”
Answering from your own documents
AI-facilitatedTalk Sam out of training the handbook in
Sam thinks “we need to train the model on our data”
Answering from your own documents
Echo replayReplay: it is just searching for matching words
an analyst at a retailer thinks “it searches for matching words”
Asking a table a question
AI-facilitatedTalk Lee out of hidden rows still counting
Lee thinks “a filter hides rows but they still count”
Kept because they landed — a peer’s wording often beats the tutor’s.
“Don't train it. Look the page up at the moment of asking and put it in front of the model. Training takes hours and is already wrong if the policy moved this morning. Lookup takes milliseconds.”
54 found this useful
“Unknown isn't false. The filter only keeps a row when the test is definitely true. A blank channel isn't 'not email' — it isn't anything. That's why 4,000 people dropped out of our campaign report.”
48 found this useful
“Forty names in that stack. Which one should it show you? Once they're stacked, the only things with one clear value are what you grouped by and the number you worked out.”
36 found this useful
“WHERE channel != 'email' — that's the same trap as the 4,000. I'd reject this. The total will go quietly small and nothing will error.”
34 found this useful