Notices by Jeremy Kahn (trochee@dair-community.social)

@blogdiva

I think that the US relationship to race and immigration is the unholy synthesis of chattel slavery (via the British-dominated triangle trade) and La Conquista peonage, which is the outgrowth of Euro colonial adventures in (North) Africa

And when Europeans say "we're not racist like Americans" they usually are focused on *one* of those two but are willfully blind to the other

French liberals overlook their history in Algiers; Belgians overlook Congo; Italians Ethiopia

@futurebird

This is making me THINK, as you and your formicogical thoughts often do.

I love this analogy; what is a mirror to a colony organism, anyway: an optical mirror is meaningless to a hive mind

But a massive LLM is like a papier-mache collage from our collective written output

it's a mirror to the culture (collective organism?) looking backwards in time

And we react to it like the cats that puff up in the mirror at that weird other interloper

https://huggingface.co/spaces/society-ethics/annotated-encyclical

Meg Mitchell, aka Shmargaret Shmitchell, with whom I am proud to be acquainted, has provided an annotation of an AI paper by none other than the Pope

The world is a strange and mysterious place

Every paragraph rings true here, at least WRT my experience at work in the last six months.

We've killed a lot of sparrows.

The AI Great Leap Forward
https://leehanchung.github.io/blogs/2026/04/05/the-ai-great-leap-forward/

This study's results suggest that LLM sycophancy amplifies publication bias (the "file drawer" problem)

The LLMs didn't like to tell the user that their hypothesis was not confirmed by the data

H/t @bibliolater
https://social.vivaldi.net/@bibliolater/116252085690032777

@inthehands

You have put your finger on the weak spot in all the hype:

Mistaking a sigmoid for an exponential

It's behind every bubble and pyramid scheme and End Times Prophecy; variations include whether the speaker themselves believes in the exponential

a professional I work with uses this very weird header image that is EITHER
* very clever
* rogue lorem ipsum that speaks to very poor awareness of web design

and while I'm afraid to know the answer

I kinda want to steal the "very clever" for my own non-existent consulting company

the X gives it a Quentin-Quire vibe that I really like

@inthehands

Jesus.

@inthehands

your "extremist" position reminds me of this short story (from @garbados ?) that i have had open in a spare tab for *months*

https://blog.bovid.space/they-let-me-live

it is oddly hopeful about the aftermath of adopting the extremist position

"marshmallow testing" isn't testing "grit"

it's a test for learned precarity: whether the child feels like they can trust the promises of adults.

@cthos

much love for

> It's worth pointing out that you, writing good unit tests and utilizing static code analysis (linters, anyone?) should be able to catch the same amount of stuff that an LLM would, deterministically.

@cthos @xgranade

and regardless of whose thought this was

I came across a case where a co-worker was laboriously writing an "agent script" for a tedious and predictable case that should really have just been procedurally generated framework code (with a variable for "what language is this")

@inthehands

I had a conversation with the head of my kid's school about AI

I pointed out that _when an AI is good at a task_ that is usually a _condemnation of the task_ ("why show me a 350-word essay when you could just show me the 40-word prompt instead?")

yours is the same insight, but for music

Never have I seen a more pointed example of "generative AI is very good at cliché"

https://www.theguardian.com/technology/2025/jul/14/an-ai-generated-band-got-1m-plays-on-spotify-now-music-insiders-say-listeners-should-be-warned

@xgranade @emma

> The Gavin N of media companies

Devastating

wait — somebody SHOT the "change my mind" guy ?

@xgranade 😱 oh boy here we go; is this going to be somebody's casus belli

VP wants examples of how we are "5x-ing our capacity" with AI, in a remarkable display of top-down wishful thinking

even supposing this were possible

any projects that are currently 5x beyond our capacity are _not being done_

Followup:

hero co-worker has used an internal LLM to generate 10 examples of how we might do this

[the examples are so vapid y'all, I am laughing so hard i'm almost crying]

coworker: "i'm hoping he notices so i can use it to make a point"

me, in response:
"But what if he doesn't notice?"