Week 0 Foundation, Tuesday, September 15
A free hour on what these tools actually are.
Sixty minutes, online. What a language model is, what an agent is, and whether you are allowed to use it at work. Nothing to prepare and no account needed.
What this hour covers
It covers what a language model actually is and how it learned what it knows, what an agent is and how the two relate, where each one earns its place in the work you already do, how to find out what your own account allows, and the three places people actually start.
Tuesday, September 15, noon Pacific. Free, and nothing to prepare.
What you will learn to do in this hour
- Interview me
- Make it ask you what it needs before it answers.
Four things you can try, each about a minute
- One
Ask which parts of a document came from the material you supplied and which were worked out. Read the two lists. That is the review boundary, visible on your own material.
- Two
Add: plan the steps first, stop before executing any of them, and mark the step where a person should approve. That is the delegation boundary, specific to your business.
- Three
Identify your account type from the billing record, find the training clause for that tier, and write down what it says. That is a fact about a contract, and where you cannot find the answer, that absence is the finding.
- Four
Take the next instance of the job you rebuild every month and record four things: how often, how long it takes, who else touches it, and how many revisions before it is accepted. That is a baseline, and it is what week one records.
The four tests written out, with what each one produces and where to write the answer down. Three pages, PDF.
How to tell which account type you are on, and where the training clause for that tier is written. Four pages, PDF.
The thirteen fundamentals
Once a move has a name you can call it on work the sessions never covered, and on whichever tool you are using next year. Two land each week, three in week three.
The same thirteen written out below, in a file you can keep beside the screen. Two pages, PDF.
Before you ask
- Interview me
- Make it ask you what it needs before it answers.
- Show three
- Give it three examples of good, then ask for a fourth.
- Break it down
- Split the job into steps instead of asking for all of it at once.
Making it better
- Set the bar
- Produce it, have it write the criteria a good version has to meet, measure the draft against them, then fix it.
- Argue against it
- Have it take the other side, and say what a hostile reader hits first.
- Around the table
- Have it read the same piece as the CFO, then the skeptic, then the customer.
When it is big
- Make it a table
- Messy input in, structured rows out.
- Same columns
- Every option assessed on the same dimensions, so they can be compared.
- Only from this
- Work from the documents you supply and nothing else.
Doing it again
- Write it once
- Build the instruction once instead of re-explaining the job every time.
- Stop and check
- Chain the steps, and name the point where a person has to look.
Trusting it
- Found or figured
- Mark what came out of the document and what was concluded on top of it.
- Back to the source
- Check against the original document, never against a summary of it.
What this site never does
There is nowhere here to upload anything, and there never will be. Your material stays yours and stays where you keep it. Nothing you work on in a session enters a Kizata system: the instructions you build live in your own account, and what is written down here is a record of what you built rather than the thing itself. Work you want read comes back by email, and the rule for that is to have the output replace every figure with a placeholder before you send it.
Take a seat in cohort 1
Six Tuesdays from September 22, noon Pacific. September 22 and 29, October 6, 13, 20 and 27.
Booking closes Friday September 18.
Ten seats. Anyone past ten starts with the next cohort.