Arun Sundararajan

Workforce capacity development and occupational transitions with dignity Arun Sundararajan explores how digital disruption is transforming work and calls for new systems to support a post-employment economy.

Potential legislative solutions to this challenge include the Retirement Savings for Americans Act of 2023 and the Portable Benefits for Independent Workers Pilot Program Act.

@digitalarun.bsky.social describes these and others in more detail in the article.

www.brookings.edu/articles/wor...

The History of Impoundments Before the Impoundment Control Act of 1974 | The Center for Renewing America Defenders of the Impoundment Control Act misunderstand the power of the purse, overread Supreme Court precedent, and fail to address unbroken Executive impoundment practice and congressional acquiesce...

a long article about whether the “power of the purse” that Congress wields is a floor or a ceiling on Presidential spending.

How My Trip to Quit Sugar Quickly Became a Journey Into Hell For my whole life, I’ve been a hard-core sweets junkie. Could a spa help me quit in a week?

a fun (long) read from @caity.bsky.social. As Kramer once said, it’s “a story about love, deception, greed, lust and unbridled enthusiasm. yes, that’s what led to Billy Mumphrey’s downfall…”

Our first Tech Leadership Lab of 2025 is on Wednesday 02/05 at 4:45PM: former Etsy CEO Chad Dickerson chats with @digitalarun.bsky.social about guiding Etsy from startup to IPO and beyond, growing revenue 14x 2011-2017, and lessons since imparted to dozens of startups. Sign up: bit.ly/dickersonTLL

I wonder what the intersection between people downloading VPNs for the first time specifically to get on TikTok and people vulnerable to a VPN scam is 😊

Opinion | How Democrats Drove Silicon Valley Into Trump’s Arms Marc Andreessen explains the newest faction of conservatism.

How? By shifting youth thinking from “Capitalism is good and an enabler of the good society” to “Capitalism is evil and should be torn down.” You might disagree with him and see this as a narrative fitted retrospectively to justify today’s position — but its a fascinating perspective nevertheless.

Statement by President Biden on the Executive Order on Advancing U.S. Leadership in Artificial Intelligence Infrastructure | The White House The United States leads the world at the frontier of artificial intelligence (AI). Cutting-edge AI will have profound implications for national security

Before we compete with AI for our jobs, we’ll be competing with it for our electricity. This issue will be central to the AI policy debate.
Infrastructure power needs are also going to be a critical constraint for many countries seeking AI sovereignty.

Our next Fireside Chat features Markus Herrmann Chen of the China Macro Group. Join us Monday, January 27th at 12:30PM to discuss the #TikTokBan and what the incoming Trump administration could mean for China's tech sector and US-China tech competition.

Register now: bit.ly/mhchen

Absolutely — any legislation that requires “taking the training data out” seems pointless. The transaction costs of opt-out should also be considered. Personal data might be a better candidate than copyrighted data for opt-out — we don’t let copyright holders opt out of other forms of fair use.

Mark Zuckerberg gave Meta's Llama team the OK to train on copyrighted works, filing claims | TechCrunch Meta CEO Mark Zuckerberg gave Meta's Llama team approval to train on copyrighted documents, according to a new court filing.

The week’s more visible changes in Meta content filtering obscured these revelations about their training data governance choices. Here’s hoping 2025 brings greater clarity about how broadly the #fairuse doctrine covers copyrighted works being used to train #AI systems like ChatGPT, Claude and Llama

if they do acquiesce to an acquisition, wouldn’t Google/Waymo be another possible suitor? Apart from their user base, Uber and Lyft have deep operational expertise and IP, something that Waymo or Google can’t rapidly replicate, and which seems very synergistic with their existing tech.

Related: an excellent 1/1 NYT oped by Jessica Grose about the virtues of modest levels of solitude, and of “embracing loneliness as a normal feeling that we all experience from time to time, rather than necessarily being pathological.”

“Idleness as such is by no means a root of evil; quite the contrary, it is a truly divine way of life, so long as one is not bored,” Kierkegaard wrote.

the #OpenAI plan to become a for-profit company reminds me of this old cartoon.

3/ Is it sufficient to accept superficial fixes that balance out generated content for a vast majority of prompts (like the LLMs and image generators have done) or do we need to delve deeper into understanding the machine equivalent of the gender beliefs of their underlying models?

Evaluating LLMs for Gender Disparities in Notable Persons This study examines the use of Large Language Models (LLMs) for retrieving factual information, addressing concerns over their propensity to produce factually incorrect "hallucinated" responses or to ...

2/ What’s the right metric to measure bias in generative AI models? Do we want them to reflect “base rate” reality, or do we want them to be aspirational? Two colleagues and I looked into this a little a year or so ago in the context of LLM recall of Nobel Prize winners.

Sora has a bias problem Sora seems to think all academics are men, and predominantly white men at that. And this is a problem.

1/ in an interesting experiment, #sora created a white male-presenting character when prompted to generate an “academic giving a lecture.” what’s surprising is that it consistently did so on 16 consecutive trials. This will likely be “rectified” rapidly, but raises familiar (and hard) questions:

I'm hiring a one-year post doc for research on AI, new tech and Future of Work

Ideal candidate has Special Sworn Status and Census RDC experience.

Interfolio link: apply.interfolio.com/158246
(NYU's term for "post doc" is "research scientist")

AEA JOE link:
www.aeaweb.org/joe/listing....

Did OpenAI Just Solve Abstract Reasoning? OpenAI’s o3 model aces the "Abstraction and Reasoning Corpus" — but what does it mean?

Some of my thoughts on OpenAI's o3 and the ARC-AGI benchmark

aiguide.substack.com/p/did-openai...

Arun Sundararajan on LinkedIn: Apart from feeling the usual good cheer, this Christmas Day I’ve been… Apart from feeling the usual good cheer, this Christmas Day I’ve been thinking about one of my favorite papers in grad school, “The Deadweight Loss of…

Triggered by @annagoeddeke.bsky.social's post: "The Deadweight Loss of Christmas" (a very cool paper) makes me wonder if government economic policy bodies should include experts from the other social sciences with an established track record of interdisciplinary thinking. Longer version below.

'The Hunt for Red October' is a Christmas movie Dive with me into a post that will submarine your 'Die Hard' theory

THE HUNT FOR RED OCTOBER is a Christmas movie. Change my mind.

YouTube video by TEDx Talks Can we own our intelligence in the age of AI? | Arun Sundararajan | TEDxGateway

18 months later, my June 2023 TEDx talk about the challenges that generative AI might pose to human intellectual autonomy is finally online. Would be delighted to hear what you think.
(And yes, I know, Google’s LLM is not called Bard any more:)

What’s something you didn’t do in your 20s that you wish you had done? (Granted, this works better for groups in which there’s a critical mass of people willing to admit not currently being in their 20s, but it engages the younger folks in a mixed-age group a lot more than one might expect.)

I’m enjoying Wolfgang Streeck’s “Taking Back Control.” I don’t agree with a lot of it, but it’s an interesting read.

Opinion | Bluesky Is Different From X. For Now. Users are looking for a more curated experience on a platform that is not constantly trying to get them to interact with brands.

I wrote something about the migration from X to Bluesky. (I did not choose the title, and do not like it, but there you go.)

www.nytimes.com/2024/12/07/o...

holding a shiny round silicon wafer

me holding a silicon wafer at ASML’s headquarters last week. i’d argue that their machines represent humanity’s most incredible engineering accomplishment to date.

2/ True, some are vertically integrated (Google, maybe Meta), but OpenAI and Anthropic aren’t, and all depend on Nvidia (and thus indirectly on TSMC and ASML). The hardware-software value capture dynamic could also invert. Many economic forces counter data-driven winner-take-all dynamics from LLMs.

1/ Looking at the premise of the original article linked to — I’m not sure that the “LLM builders” today are going to be the eventual value capture layer. They are sandwiched between the user layer (copilot, retail, advertising) and the infrastructure layer, both of which seem to have more power.

this is very compelling. i don’t think we really have a good understanding of what has led to the sudden dramatic recent success of fantasists. Part of this might be understanding what has made humans susceptible to them in the past, and perhaps also how fragmented the “chorus” was pre-broadcast TV.

Artificial Intelligence, Human Intellectual Autonomy and the Future of Work In January 2015, Congress convened the inaugural meeting of its newly-formed Sharing Economy Caucus. Led by Representatives Jerome Nadler and Eric Swalwell, a group of lawmakers, tech CEOs, and a few ...

so here’s an essay i wrote for the @aspeninstitute.bsky.social recently. our focus on copyright and training data is making us miss the more important IP challenge raised by #AI: making sure people have sufficient ownership over their human capital. www.aspeninstitute.org/blog-posts/a...