Lightnews — Scholar-powered news

Reposted by conputer dipshit

Amanda Katz @katzish.bsky.social · 3h

Horrific reporting from the AP on the Myanmar families watching their kids starve to death from Trump and Musk’s aid cuts apnews.com/article/myan...

Starving children screaming for food as US aid cuts unleash devastation and death across Myanmar

U.S. Secretary of State Marco Rubio has repeatedly said “no one has died" because of his government’s decision to gut its foreign aid program.

apnews.com

21 610 970

Reposted by conputer dipshit

Henry Farrell @himself.bsky.social · 2h

"It wanted to signal strength. Instead, it’s revealing its weakness. The administration’s need to break the academy is forcing it to make a desperately risky gamble." www.nytimes.com/2025/10/08/o...

Opinion | You Beat Trumpism by Banding Together. It’s as Hard and as Simple as That.

www.nytimes.com

5 70 260

conputer dipshit @davidcrespo.bsky.social · 15h

and he's obviously not the only one

conputer dipshit @davidcrespo.bsky.social · 7d

I don't know what to say other than that this is a grave factual error on which to base your analysis. it's fucking ridiculous

the AI bubble is driven by monopolists who've conquered their markets and have no more growth potential

graph of msft, meta, goog, and amzn revenue over time

1

conputer dipshit @davidcrespo.bsky.social · 15h

we have the receipts

conputer dipshit @davidcrespo.bsky.social · Sep 4

I wrote about Ed Zitron using words to mean the opposite of what they normally do because it makes things sound more dire for the LLM industry. this is not a good method if you want to think clearly and understand things
crespo.business/posts/cost-o...

Is the “cost of inference” going up or down? – David Crespo

TVs have been getting cheaper for decades, yet people are spending more on them than ever. Does anyone find this puzzling? No, it’s obvious how this happens: th…

crespo.business

1 2

Reposted by conputer dipshit

Eli Valley @elivalley.bsky.social · 19h

Bari Weiss, CBS News, 1965

- New art in @jewishcurrents.bsky.social
jewishcurrents.org/bari-weiss-c...

26 680 3.1K

Reposted by conputer dipshit

Berny Belvedere @bernybelvedere.bsky.social · 23h

fellas is it gay to live in a city

260 310 3.3K

Reposted by conputer dipshit

Chris Geidner @chrisgeidner.bsky.social · 21h

First take out of arguments: Despite efforts by Sotomayor to put off a decision, it appeared the Supreme Court will subject Colorado’s conversion therapy ban to strict scrutiny under the First Amendment, with a majority likely to strike it down outright.

More to come at Law Dork:

Law Dork | Chris Geidner | Substack

The Supreme Court, law, politics, and more. Click to read Law Dork, by Chris Geidner, a Substack publication with tens of thousands of subscribers.

www.lawdork.com

9 41 96

conputer dipshit @davidcrespo.bsky.social · 21h

in some cases at least, I feel like I'm seeing it get easier to make the case for these things when the LLM can't work without them. like it seems like people are improving their docs and API references somewhat

Reposted by conputer dipshit

Tamoor Hussain @tamoorh.com · 1d

Ok this slaps

3 78 360

conputer dipshit @davidcrespo.bsky.social · 22h

and even if they do become better than people at that, as you say they will be directed to more effectively build evil things. so for an appropriately general sense of "bad software", no they will not reduce it

1 1

conputer dipshit @davidcrespo.bsky.social · 22h

not sure — depends on how much LLMs can help on the judgment side. so far I think SOTA LLMs are better than the average human engineer at reviewing a PR for mistakes, possibly better than a good human engineer. not so for "am I building the right thing," that are probably more important overall

1 1

conputer dipshit @davidcrespo.bsky.social · 22h

really really good

1

Reposted by conputer dipshit

whet moser @whetmoser.com · 23h

this is really good

Sarah Kaplan @sarahkaplan48.bsky.social · 23h

An extraordinary story made possible by an unprecedented partnership between Washington Post reporters and readers. It takes immense trust to give a journalist access to your data and your story. My colleagues and I strive to earn and keep that trust every day.
www.washingtonpost.com/wellness/int...

How TikTok keeps its users scrolling for hours a day

Washington Post reporters analyzed data from over 800 TikTok users to learn how the app turns people into power users, some of whom spend hours per day scrolling.

www.washingtonpost.com

1 2 6

conputer dipshit @davidcrespo.bsky.social · 23h

which is Simon's point. these have been the conditions of high quality software engineering for a long time before LLM tooling came out

Automated testing. If your project has a robust, comprehensive and stable test suite agentic coding tools can fly with it. Without tests? Your agent might claim something works without having actually tested it at all, plus any new change could break an unrelated feature without you realizing it. Test-first development is particularly effective with agents that can iterate in a loop.
Planning in advance. Sitting down to hack something together goes much better if you start with a high level plan. Working with an agent makes this even more important—you can iterate on the plan first, then hand it off to the agent to write the code.
Comprehensive documentation. Just like human programmers, an LLM can only keep a subset of the codebase in its context at once. Being able to feed in relevant documentation lets it use APIs from other areas without reading the code first. Write good documentation first and the model may be able to build the matching implementation from that input alone.
Good version control habits. Being able to undo mistakes and understand when and how something was changed is even more important when a coding agent might have made the changes. LLMs are also fiercely competent at Git—they can navigate the history themselves to track down the origin of bugs, and they’re better than most developers at using git bisect. Use that to your advantage.
Having effective automation in place. Continuous integration, automated formatting and linting, continuous deployment to a preview environment—all things that agentic coding tools can benefit from too. LLMs make writing quick automation scripts easier as well, which can help them then repeat tasks accurately and consistently next time.
A culture of code review. This one explains itself. If you’re fast and productive at code review you’re going to have a much better time working with LLMs than if you’d rather write code yourself than review the same thing written by someone (or something) el…

2 3

conputer dipshit @davidcrespo.bsky.social · 23h

it's very important to articulate how to do it well, but I don't know if we need a term for it — at the current rate of adoption, in a few months it'll just be "software engineering"

Simon Willison @simonwillison.net · 23h

Vibe coding is irresponsibly building software through dice rolls, not caring what code is produced

What about when engineers at the top of their game use AI tools responsibly to accelerate their work?

I propose "vibe engineering"!

simonwillison.net/2025/Oct/7/v...

Vibe engineering

I feel like vibe coding is pretty well established now as covering the fast, loose and irresponsible way of building software with AI—entirely prompt-driven, and with no attention paid to …

simonwillison.net

1 1 2

conputer dipshit @davidcrespo.bsky.social · 1d

only artist that comes to mind is joanna newsom but I'm sure I could come up with a few others I like as much

Emily, a Two-Shoes-Gooder @rebelmusicteach.bsky.social · 2d

I also need to know: I am fairly middle-of-the-road on Taylor Swift but someone I know called her the Greatest Living Lyricist. She is not.

Who is yours? Either the greatest or your favorite?

2

conputer dipshit @davidcrespo.bsky.social · 1d

exactly

1

conputer dipshit @davidcrespo.bsky.social · 1d

I won't pour myself a glass of milk but if one of my kids leaves half their cup I will enjoy it. ideally at room temp

Faine Greenwood @faineg.bsky.social · 1d

What foods do you love that you fully acknowledge make you a pervert for loving them?

2

conputer dipshit @davidcrespo.bsky.social · 1d

midnight pickle guys rise up

1 1

conputer dipshit @davidcrespo.bsky.social · 1d

😐

1

conputer dipshit @davidcrespo.bsky.social · 1d

they're pretending to be worked up about this while the president is openly directing the DOJ to invent criminal charges for a list of his enemies on truth social

screenshot of cernovich QTing chuck grassley

@Cernovich This is something out of East Germany.

@ChuckGrassley This document shows the Biden FBI spied on 8 of my Republican Senate colleagues during its Arctic Frost investigation into "election conspiracy" Arctic Frost later became Jack Smith's elector case against Trump

1 1

conputer dipshit @davidcrespo.bsky.social · 1d

weird feeling reading the judge's order and thinking "wow! there is a person in power for whom facts matter"
www.portland.gov/federal/docu...

In sum, the President is certainly entitled “a great level of deference,” Newsom II, 141
F.4th at 1048, in his determination that he “is unable with the regular forces to execute the laws
of the United States.” 10 U.S.C. § 12406(3). But “a great level of deference” is not equivalent to
ignoring the facts on the ground. As the Ninth Circuit articulated, courts must “review the
President’s determination to ensure that it reflects a colorable assessment of the facts and law
within a ‘range of honest judgment.’” Id. at 1051 (quoting Sterling, 278 U.S. at 399). Here, this
Court concludes that the President did not have a “colorable basis” to invoke § 12406(3) to
federalize the National Guard because the situation on the ground belied an inability of federal
law enforcement officers to execute federal law. Id. at 1051–52. The President’s determination
was simply untethered to the facts.

3

Reposted by conputer dipshit

hyperplanes.bsky.social @hyperplanes.bsky.social · 1d

that thing everyone said was inevitable happened bsky.app/profile/jsra...

John Scott-Railton @jsrailton.bsky.social · 2d

NEW: breach of Discord age verification data.

Including some users passports & DLs

Age verification is a badly implemented data grab wrapped in a moral panic.

Mark my words, as age verification mandates expand, we'll end up more surveilled and less secure. 1/

2 22 42

conputer dipshit @davidcrespo.bsky.social · 1d

(in that chart and the text, OpenAI are being sloppy mixing up logged-in and total WAUs though)

1

conputer dipshit @davidcrespo.bsky.social · 1d

rare useful citation from Zitron, though of course if you read it, it proves him wrong: overestimate due to double-counted logged-out users is probably pretty small, and even if you revise all stats downward by 10% it doesn't change the rate on the exponential curve
cdn.openai.com/pdf/a253471f...

‪Ed Zitron‬
‪@edzitron.com‬
· 2h
1. 800 million weekly actives, but as we found out from OpenAI's study, they don't attempt to deduplicate logged-out users.
2. They said 3 million developers "built with OpenAI" a year ago, is 4 million really that great, especially considering GPT-5?

techcrunch.com/2024/10/01/o...
As OpenAI undergoes yet another C-suite overhaul – a reminder of the turmoil following last year’s DevDay – the company is trying to convince developers that it still offers the best platform to build AI apps on. Leaders say the startup has more than 3 million developers building with its AI models, but OpenAI is operating in an increasingly competitive space.
ALT
‪Techmeme‬
‪@techmeme.com‬
· 4h
Sam Altman says ChatGPT has reached 800M weekly active users, 4M developers "have built with OpenAI", and OpenAI processes over 6B tokens per minute on its API (Rebecca Bellan/TechCrunch)

Main Link | Techmeme Permalink
Ed Zitron
‪@edzitron.com‬
www.nber.org/system/files...
20Note that we expect our counts of distinct accounts to somewhat exceed distinct people when one person has two accounts (or, for logged-out users, one person using two devices). For logged-in users, the count is based on distinct login credentials (email addresses), and one person may have multiple accounts. For logged-out users, the count is based on distinct browser cookies; this would double-count people if someone returns to ChatGPT after clearing their cookies, or if they access ChatGPT with two different devices in the same week.

4 The Growth of ChatGPT
ChatGPT was released to the public on November 30, 2022 as a “research preview,” and by December
5 it had more than one million registered users. Figure 3 reports the growth of overall weekly active
users (WAU) on consumer plans over time. ChatGPT had more than 100 million logged-in WAU after
one year, and almost 350 million after two years. By the end of July 2025, ChatGPT had more than
700 million total WAU, nearly 10% of the world’s adult population.20
Figure 3: Weekly active ChatGPT users on consumer plans (Free, Plus, Pro), shown as point-in-time
snapshots every six months, November 2022–September 2025.
Figure 4 presents growth in the total messages sent by users over time. The solid line shows that
between July 2024 and July 2025, the number of messages sent grew by a factor of more than 5.
Figure 4 also shows the contribution of individual cohorts of users to aggregate message volume.
The yellow line represents the first cohort of ChatGPT users: their usage declined somewhat over
2023, but started growing again in late 2024 and is now higher than it has ever been. The pink line
represents messages from users who signed up in Q3 of 2023 or earlier, and so the difference between
20Note that we expect our counts of distinct accounts to somewhat exceed distinct people when one person has two
accounts (or, for logged-out users, one person using two devices). For logged-in users, the count is based on distinct
login credentials (email addresses), and one person may have multiple accounts. For logged-out users, the count is based
on distinct browser cookies; this would double-count people if someone returns to ChatGPT after clearing their cookies,
or if they access ChatGPT with two different devices in the same week.

1 3