David Crawford
banner
david-crawford.bsky.social
David Crawford
@david-crawford.bsky.social
“Data informs. Logic guides. Decisions follow.”
Optimizing and/or exploring. New York, New England, Montreal.
Reposted by David Crawford
My thanks to Holly Krambeck for inviting me to keynote at the World Bank’s Data Partnership Day (and sending this picture!), discussing data, policy, and 'How Data Happened' ( https://bit.ly/hdh-book ), w/ Hal Varian of Google
& Karin Kimbrough of LinkedIn
May 30, 2023 at 5:17 PM
1) Report what they did
2) report what they observed
3) recite the standard procedure and
4) minor paraphrasing to reconcile standard procedure (3) with what they said they did (1).

Explaining why standard procedure is necessary, sufficient, and reliable: beyond the scope of those who follow it.
April 20, 2025 at 1:14 AM
METR access early checkpoints, technical information about o3 and o4-mini, a subset of the model’s benchmark results from OpenAI internal evaluations. METR evaluated model’s performance on its task suites for general autonomy (HCAST) and AI R&D (RE-Bench). metr.github.io/autonomy-eva...
Details about METR’s preliminary evaluation of OpenAI’s o3 and o4-mini
Resources for testing dangerous autonomous capabilities in frontier models
metr.github.io
April 18, 2025 at 2:40 AM
Excellent taxonomy of critics and opponents of AI by @benjaminjriley.bsky.social
buildcognitiveresonance.substack.com/p/who-and-wh...

1) Scientific Skeptics – Cognitive Science Wing
believes generative AI based on Large Language Models is limited in its capabilities compared to human cognition
April 9, 2025 at 12:05 AM
Reposted by David Crawford
The endless back and forth over AGI is getting stale. We need better ways to talk about what people should do when they use large AI models. @alisongopnik.bsky.social @himself.bsky.social @dsquareddigest.bsky.social offer that and more.

On Beyond AGI! www.ailog.blog/p/on-beyond-...
On Beyond AGI!
We need new language for talking about large AI models. Alison Gopnik, Henry Farrell, James Evans, Cosma Shalizi, and Dan Davies provide one.
www.ailog.blog
April 6, 2025 at 11:05 AM
Just tell me what to do. #UX
April 2, 2025 at 2:41 AM
Nominative determinism

Philip Langstaffe Ord Guy aka PLO Guy became a Palestine Mandate bureaucrat.
I’ll never stop chuckling when I’m reminded of it. A British army officer who was involved in several archeological digs around the Middle East in the interwar period and then became a key archeological bureaucrat of sorts during the Palestine Mandate period was named P. L. O. Guy.
March 17, 2025 at 9:10 PM
Also, developers calling most new developments #luxury is a fraud.

The term “luxury” has traditionally denoted exclusivity, superior craftsmanship, and unparalleled amenities, as seen in grand estates, private penthouses, and homes in the most prestigious addresses. 1/5
Yes, decades of strict zoning, design review, and historic preservation have created a disastrous housing shortage that turned every home into luxury housing; so the city is changing the rules to make it easier for developers to invest in building more homes for people to live in.
March 3, 2025 at 11:56 PM
Reposted by David Crawford
"and by the time it reaches everybody, millions of people have seen it. "

A tautology of attenuation and ablation.
Watters: We are waging a 21st century information warfare campaign from the left…It's like grassroots guerrilla warfare. Someone says something on social media, Musk retweets it, Rogan podcasts it, Fox broadcasts it.. and by the time it reaches everybody, millions of people have seen it.
February 18, 2025 at 9:53 PM
Is there a preferred way to save images with their associated alt text so that when fresh posting or posting on another platform, the original alt text is retained ?

Also I assume that in some cases that alt text may make an image, and any doc/post/tweet/etc, containing the image, more findable.
January 29, 2025 at 9:17 PM
Reposted by David Crawford
“This time is different. With older media, the friction of the interface provided some space for reflection and hierarchizing significance. What was on the front pages or what led the news bulletins was what we heeded most.” Now fragmentation reigns.
lareviewofbooks.org/article/the-...
The Case for Kicking the Stone | Los Angeles Review of Books
Philip Ball finds Nicholas Carr’s “Superbloom: How Technologies of Connection Tear Us Apart” disturbingly compelling.
lareviewofbooks.org
January 29, 2025 at 2:01 AM
Reposted by David Crawford
AI “made it easier for scammers and spammers to generate fake jobs and fake applications. At the same time, real job seekers felt pressure to keep up with the application numbers they saw others posting on social media”
thebaffler.com/outbursts/hi...
Hiring Squad | Megan Marz
Job candidates are now routinely filtered and sorted by algorithms. But a total panopticon awaits workers once they’ve been hired.
thebaffler.com
January 25, 2025 at 6:09 PM
Time to focus-group cleanse vs purge ? #SLATE_PITCH @nytpitchbot.bsky.social
It's notable to me that NY Times actually uses "purge" here.

www.nytimes.com/2025/01/24/u...
January 25, 2025 at 8:27 PM
Mass-produce new visions.
Large language model (LLM) hallucinations are "a potential resource instead of a categorically negative pitfall" arxiv.org/pdf/2406.04175 hmm
January 25, 2025 at 8:17 PM
Reposted by David Crawford
Is there a metadata or image format that saves or otherwise retains the application layer Alt-text provided with images ?
People aggressively scolding me re: alt text never worked as they hoped. But @rahaeli.bsky.social did advocacy right, with patience and empathy.

So right, in fact, that I was actually *excited* to find a new Brave Browser feature that extracts text from an image right from the right click menu.
January 22, 2025 at 5:42 PM
Reposted by David Crawford
This is a figure from a new paper on "The Function of Memes in Political Discourse" pmc.ncbi.nlm.nih.gov/articles/PMC... Warning - the full paper contains a skibidi toilet
January 14, 2025 at 4:10 PM
Reposted by David Crawford
The @bencasselman.bsky.social article about economists’ influence is good and you should read it. Some thoughts inspired by it and reactions I’ve seen (potentially interrupted by meetings)… www.nytimes.com/2025/01/10/b...
Economists Are in the Wilderness. Can They Find a Way Back to Influence? (Gift Article)
Economists have long helped to shape policy on issues like taxes and health care. But flawed forecasts and arcane language have cost them credibility.
www.nytimes.com
January 10, 2025 at 5:59 PM
Controlling the citizen, an ongoing saga.
January 11, 2025 at 12:49 PM
A headline 'Moving average calculation costs renters $3.4 billion' would get fewer clicks. Need #AI for rizz.
San Francisco has a ban on algorithmic rent setting software, a vacancy tax, rent control, and strong tenant protections.

As a result, it’s *checks notes* America’s fourth most expensive city.
AI software owned by a private equity firm costs U.S. renters an extra $3.6 billion per year, artificially driving up rents and increasing housing insecurity.

We criminalized homelessness when we should’ve criminalized price gouging housing instead.
popular.info/p/ai-costs-a...
January 7, 2025 at 3:41 AM
Reposted by David Crawford
"I don't buy the idea that GenAi is Armageddon for credible information and an informed public. I have a hard time seeing the evidence for that". Most people operate with healthy skepticism and are capable of navigating media environment - spoke to @kforum.bsky.social kforum.dk/nyheder/fors...
Professor: "Jeg køber ikke ideen om, at AI er en dommedag for troværdig information"
Generativ AI er det vigtigste emne med relevans for kommunikation i 2025, lyder det fra Rasmus Kleis Nielsen, professor i kommunikation.
kforum.dk
January 6, 2025 at 9:17 AM