The computer is a data knife. I've been using computers to cut up data since I first got one with enough memory to hold data in the late 1980s. The ergonomics of doing this, however, have been uniformly poor the whole time. Spreadsheets are fine for a certain subset of data tasks, kind of like a bagel-cutter is good for slicing a certain subset of foods. Most database software is structurally industrial-scale, more like a bread-slicing machine than a knife.
A friend just sent me this tiny table of numbers:
$100 25 shares
$500 50 shares
$1K 100 shares
$5K 250 shares
$10K 500 shares
$50K 1,000 shares
$100K 2,000 shares
$500K 5,000 shares
$1M 10,000 shares
$5M 25,000 shares
$10M 50,000 shares
$50M 100,000 shares
$100M 250,000 shares
$500M 500,000 shares
$1B+ 1,000,000 shares
I pasted it into Google Sheets, did Data > Split text to columns, and got this:
Not useful.
But do we still need data knives if we have AI? Our agents can cut our data up into safe bite-sized pieces for us, can't they? Hold still and open your mouth. I gave this table to Claude and asked "does this look right?"
My questions there are verbatim, Claude's responses are condensed from multiple pages of earnest unhelpfulness. That may not be a dull knife, exactly, but I don't really want a cheerful knife with which I have to begin by arguing about the nature of sharpness.
Part of my ambition for DACTAL is that it could be a better knife. The surgery we need here isn't exactly traumatic: expand the K/M/B shorthands into actual numbers, chop up this set of space-separated-values into labeled fields, check the math, maybe try an alternative scheme.
That isn't just a picture of numbers. Click it and you'll go to the cutting board where you can hack at them yourself. Or, if you prefer, some other data of your own.
Next I'm going to need some data tupperware.
A friend just sent me this tiny table of numbers:
$100 25 shares
$500 50 shares
$1K 100 shares
$5K 250 shares
$10K 500 shares
$50K 1,000 shares
$100K 2,000 shares
$500K 5,000 shares
$1M 10,000 shares
$5M 25,000 shares
$10M 50,000 shares
$50M 100,000 shares
$100M 250,000 shares
$500M 500,000 shares
$1B+ 1,000,000 shares
I pasted it into Google Sheets, did Data > Split text to columns, and got this:
Not useful.
But do we still need data knives if we have AI? Our agents can cut our data up into safe bite-sized pieces for us, can't they? Hold still and open your mouth. I gave this table to Claude and asked "does this look right?"
Claude: The scaling seems quite aggressive - the share awards increase much faster than the earnings.
glenn: You say "the share awards increase much faster than the earnings", but isn't it exactly the reverse?
Claude: You're absolutely right - I made an error in my analysis. The share awards increase much slower than the earnings, not faster. This creates a regressive structure where higher earners get a smaller percentage of their earnings in shares.
glenn: If higher earners get a smaller percentage, isn't that literally a progressive scheme?
Claude: You're absolutely correct again - I have my definitions backwards!
My questions there are verbatim, Claude's responses are condensed from multiple pages of earnest unhelpfulness. That may not be a dull knife, exactly, but I don't really want a cheerful knife with which I have to begin by arguing about the nature of sharpness.
Part of my ambition for DACTAL is that it could be a better knife. The surgery we need here isn't exactly traumatic: expand the K/M/B shorthands into actual numbers, chop up this set of space-separated-values into labeled fields, check the math, maybe try an alternative scheme.
That isn't just a picture of numbers. Click it and you'll go to the cutting board where you can hack at them yourself. Or, if you prefer, some other data of your own.
Next I'm going to need some data tupperware.