conputer dipshit
@davidcrespo.bsky.social
web dev + hot dad. enjoy charts, unions, conputer games, philosophy. chicago crespo.business
everything that uses a boring ass regression or a decision tree is AI now
November 11, 2025 at 1:50 AM
everything that uses a boring ass regression or a decision tree is AI now
I mean plenty of famous good devs are saying it: Mitchell Hashimoto, Armin Ronacher, etc
November 11, 2025 at 12:39 AM
I mean plenty of famous good devs are saying it: Mitchell Hashimoto, Armin Ronacher, etc
right, it's like if a thing is changing noticeably every two months, you should probably check in at least every six months?
November 10, 2025 at 11:13 PM
right, it's like if a thing is changing noticeably every two months, you should probably check in at least every six months?
I guess those two are compatible — it is cringe precisely because it is in the marketing
November 10, 2025 at 11:10 PM
I guess those two are compatible — it is cringe precisely because it is in the marketing
which is funny because they make it sound like bosses are holding us at gunpoint to smile and say it works
November 10, 2025 at 11:09 PM
which is funny because they make it sound like bosses are holding us at gunpoint to smile and say it works
Diarrhea Planet - "Kids" | A Shiner Session
YouTube video by Do512
www.youtube.com
November 10, 2025 at 11:06 PM
this one is a doozy
November 10, 2025 at 9:16 PM
this one is a doozy
what I'm saying is there are several axes
- in-house analyst vs. vendor product
- human vs. AI
- can fix error vs. can't fix error
all three are distinct, so you can have any combination of them
- in-house analyst vs. vendor product
- human vs. AI
- can fix error vs. can't fix error
all three are distinct, so you can have any combination of them
November 10, 2025 at 8:51 PM
what I'm saying is there are several axes
- in-house analyst vs. vendor product
- human vs. AI
- can fix error vs. can't fix error
all three are distinct, so you can have any combination of them
- in-house analyst vs. vendor product
- human vs. AI
- can fix error vs. can't fix error
all three are distinct, so you can have any combination of them
no, precisely not. because there is no guarantee the analyst can change their categorization!
November 10, 2025 at 8:48 PM
no, precisely not. because there is no guarantee the analyst can change their categorization!
I think that's too generous of an excuse for him. plenty of people who have never touched a terminal know better than he does
November 10, 2025 at 8:48 PM
I think that's too generous of an excuse for him. plenty of people who have never touched a terminal know better than he does
so much of it is cured meats. just tell people not to eat salami and bacon!
November 10, 2025 at 8:46 PM
so much of it is cured meats. just tell people not to eat salami and bacon!
the problem is control. the fact that they need to call their vendor to fix an automated error. AI is their excuse, but they could easily have said "the system miscategorized your claim" or even "an analyst miscategorized your claim" and it wouldn't change anything about the scenario
a few months ago I said "having generative AI handle absolutely anything with regards to healthcare is a nightmare and should be banned" and a bunch of people made fun of me and called me stupid. anyways,
November 10, 2025 at 8:16 PM
the problem is control. the fact that they need to call their vendor to fix an automated error. AI is their excuse, but they could easily have said "the system miscategorized your claim" or even "an analyst miscategorized your claim" and it wouldn't change anything about the scenario
the ikea episode is also great and they have a very similar approach to costco
www.acquired.fm/episodes/ikea
www.acquired.fm/episodes/ikea
IKEA | Acquired Podcast
The complete podcast (and transcript!) of IKEA’s history and business strategy.
www.acquired.fm
November 10, 2025 at 6:24 PM
the ikea episode is also great and they have a very similar approach to costco
www.acquired.fm/episodes/ikea
www.acquired.fm/episodes/ikea
I haven't listened to it yet, but every Acquired episode is solid gold
www.acquired.fm/episodes/cos...
the nintendo episodes are incredible
www.acquired.fm/episodes/nin...
www.acquired.fm/episodes/nin...
www.acquired.fm/episodes/cos...
the nintendo episodes are incredible
www.acquired.fm/episodes/nin...
www.acquired.fm/episodes/nin...
Costco | Acquired Podcast
The complete podcast (and transcript!) of Costco’s history and current business strategy.
www.acquired.fm
November 10, 2025 at 6:23 PM
I haven't listened to it yet, but every Acquired episode is solid gold
www.acquired.fm/episodes/cos...
the nintendo episodes are incredible
www.acquired.fm/episodes/nin...
www.acquired.fm/episodes/nin...
www.acquired.fm/episodes/cos...
the nintendo episodes are incredible
www.acquired.fm/episodes/nin...
www.acquired.fm/episodes/nin...
it's plausible, but there are plenty of people who aren't coders who can immediately see the use case too
November 10, 2025 at 6:07 PM
it's plausible, but there are plenty of people who aren't coders who can immediately see the use case too
so glad you asked!
they gave the caveats but here's my version:
- experts on their own open source proj are not representative of most software dev
- measuring time undervalues trading time for effort
- tools are noticeably better than they were 6 months ago
- it really does take months of use to get the hang of it
- experts on their own open source proj are not representative of most software dev
- measuring time undervalues trading time for effort
- tools are noticeably better than they were 6 months ago
- it really does take months of use to get the hang of it
November 10, 2025 at 5:53 PM
so glad you asked!
yep. to me it seemed clear very early that as long as there is *some* little bit of signal, you can do the boring engineering to make it useful
this thing about extracting non-obvious signal is the whole game IMO. it's why it was totally clear to me well before agents got good that LLMs would be good for *something*. if there's a weak signal you can build on that. was always surprised when programmers didn't see it
bsky.app/profile/davi...
bsky.app/profile/davi...
I use LLMs every day for work (programming) so to me it is way beyond doubt that it good for *something*. of course we'll how broad that will go. I think the mental model of replacing whole activities is def wrong, it's more like glue between systems that can't currently talk to each other
November 10, 2025 at 5:45 PM
yep. to me it seemed clear very early that as long as there is *some* little bit of signal, you can do the boring engineering to make it useful
I've been using them pretty much daily in a non-agentic chat CLI for 2 1/2 years. from a 2025 POV, I would describe GPT-4 as basically a piece of shit, and even then it was useful every day!
here is a useful chat I had from that time about web dev
gist.github.com/david-crespo...
here is a useful chat I had from that time about web dev
gist.github.com/david-crespo...
November 10, 2025 at 5:41 PM
I've been using them pretty much daily in a non-agentic chat CLI for 2 1/2 years. from a 2025 POV, I would describe GPT-4 as basically a piece of shit, and even then it was useful every day!
here is a useful chat I had from that time about web dev
gist.github.com/david-crespo...
here is a useful chat I had from that time about web dev
gist.github.com/david-crespo...