anthonymoser.com
(He/Him) Folk Technologist • [email protected] • N4EJ • http://www.BetterDataPortal.com • baker in The FOIA Bakery • http://publicdatatools.com/ • quick network graphs bit.ly/qng • http://deseguys.com • bit.ly/IneffectiveByChoice
5,822 posts
4,019 followers
1,631 following
Regular Contributor
Active Commenter
comment in response to
post
the hong kong protestors used the phrase "do not split" to mean that the radical protest and peaceful protest arms of the movement should refuse to criticize each other.
comment in response to
post
I guess it sort of makes Picard feel culpable but he's already in the middle of it
and even he is like well, we're just gonna play it straight like this isn't a factor
comment in response to
post
actually that may have been part of why i didn't finish that the first time, the comparison logic gets trickier when you have boundaries like this
comment in response to
post
I should go back and try it, at least for the ones that have attendance boundaries.
I did make a filterable map of historical boundaries for HS, and I remember having my assumptions violated in unexpected ways working on the data
like, I had assumed all boundaries would be contiguous
comment in response to
post
There's a common thread, I think, in the indifference of summarizing with AI, and the lack of imagination when we fail to question basic assumptions about how our schools work, and why.
Things don't have to be this way.
But to imagine something new, we have to really perceive what's in front of us
comment in response to
post
I'm pretty sure it's accurate now. But through that process I learned things about specific schools and their relationships to each other, and to the city's history.
Tonight that made me think of this excellent essay by @biancawylie.com about why it matters to spend time up close with the details
comment in response to
post
When I was working on this map, I shared some drafts on twitter and a number of folks very helpfully pointed out errors that came from edge cases.
One by one, I was able to figure out why a particular combination of shapes and boundaries was tripping up the logic of the program
comment in response to
post
to me this is connected to an important question:
why do we have attendance boundaries* at all?
why is the idea of a neighborhood school limited to just "the school you are assigned to by the district"?
(* it's capitalism and white supremacy, per usual)
comment in response to
post
you might notice that some of them include like, parks or roads, and wonder if anybody lives in these areas.
I did some followup work later using the Chicago Dots package @bunkum.us put together and tl;dr yeah, people live in them
comment in response to
post
art: it's good
comment in response to
post
An excellent point! They are surfacing associations, which these models are good at.
They are also doing that in a context that explicitly calls for human judgment
comment in response to
post
An LLM generates probable text. Sometimes, the text is also true.
What some are calling "hallucinations" can never be solved inside the model, because it describes a discrepancy *between the model and the external world*
comment in response to
post
What I said is that it has the *same relationship to truth*
If you ask yes or no questions, a coin flip will have a ~50% accuracy rate.
This does not mean the coin understands the questions
comment in response to
post
Completely agree. LLMs are like context collapse as a service
comment in response to
post
A probability map of text in a corpus can only produce "this is associated" and never "this is the answer"
The thing is "this is associated" can actually be pretty useful!
But it requires a person to critically evaluate, and it positions the tool as a support, rather than a replacement
comment in response to
post
Even if you train it entirely on vetted factual information from a specific domain, it will still either
A) produce a canned answer that may or may not be relevant
B) generate text based on the probability of tokens in the source material, which is *still structurally indifferent to truth*
comment in response to
post
I like to say they are structurally indifferent to truth
comment in response to
post
it is somewhat remarkable that he was governor first
comment in response to
post
genius is an adjective, not a noun
comment in response to
post
in person, people retain control of their own documents but online, the form and function is that you have to submit your documents for review before a city key is issued
city council should have allocated more funds to the clerk's office so she could have hired more people to do in person events
comment in response to
post
I was tangential to the Clerk's office a bit during testing and launch of CityKey via my job at the time, and I am pretty upset to learn about the December change. Such a fundamental aspect of the program, and so safety critical as this news makes clear. That tradeoff should never have been made.
comment in response to
post
*trump is elected in 2016*
Chicago: to protect non-citizens from donald trump we will make it possible to get an ID without keeping records on them
*trump is elected in 2024*
Chicago: we are going to start keeping records on them
comment in response to
post
yep, except in this case "new problems" means "directly defeating the original purpose of the program"
comment in response to
post
Chicago made a choice to create this danger.
December 2024 is, notably, after November 2024