Suresh Venkatasubramanian
LightNews LightNews
geomblog.bsky.social
Suresh Venkatasubramanian
@geomblog.bsky.social
4.3K followers 770 following 870 posts
Director, Center for Tech Responsibility@Brown. FAccT OG. AI Bill of Rights coauthor. Former tech advisor to President Biden @WHOSTP. He/him/his. Posts my own.
Posts Replies Media Videos
geomblog.bsky.social
States should not be allowed to require modification of the always truthful outputs of LLMs, sez leaked Trump EO
November 21, 2025 at 11:47 AM
geomblog.bsky.social
Serena Booth (@reniebird.bsky.social) reflects on the challenge of collecting human preferences to steer AI systems and wonders if we are doing it all wrong. cntr.brown.edu/news/2025-11...

This is part 1 of 3 of CNTR researcher reflections on COLM 2025.
November 19, 2025 at 1:36 PM
geomblog.bsky.social
To @lizz-iek.bsky.social 's point: (from moritzlaw.osu.edu/sites/defaul...)
October 16, 2025 at 7:00 PM
geomblog.bsky.social
My students know me so well.
September 9, 2025 at 7:28 PM
geomblog.bsky.social
This snippet from www.facinghistory.org/resource-lib... has been haunting me. Especially since I've seen this play out in discussions today.
August 25, 2025 at 6:35 PM
geomblog.bsky.social
I very much relate to this excerpt. I can't even listen to the radio when driving any more because within 5 minutes of putting it on someone is saying something idiotic about AI.
Most of all, I'm looking forward to AI not being on everyone's mind all the time. I want to be able to read the Economist or watch BBC and not hear about AI. No Superbowl ads either, please. After the crash, people's attention will move on to whatever the new new thing will be. Who knows, longevity drugs? Space travel? Flying electric cars? Whatever it will be, I hope it also sucks up all the people who only came to AI for the money.
August 13, 2025 at 6:10 AM
geomblog.bsky.social
I'm still contemplating this.
August 7, 2025 at 6:49 PM
geomblog.bsky.social
This is a beautiful article from @quantamagazine.bsky.social on new insights into what cells can learn and remember. www.quantamagazine.org/what-can-a-c...

But I was amused by the unintentional subtext here about our AI publishing present and why no one can remember papers written 5 years ago.
“If the world is changing really fast, you should forget things [more easily], because the things that you learned are going to have a shorter shelf life,” Gershman said. “They’re not going to be as useful later on, because the world will have changed.” These dynamics are as relevant to a cell’s existence as they are to ours.
August 4, 2025 at 5:33 PM
geomblog.bsky.social
Left prompt: "show three little boys enjoying their birthday with friends". Right: "convert this to caricature and make the boys south Asian"

I am going to wish people "Hippy berritay" from now on.
August 3, 2025 at 8:40 PM
geomblog.bsky.social
I very much like this idea. Consider this vision:
I would propose a simple rule: no answers from nowhere. This rule is less convenient, and that’s the point. The chatbot should be a conduit for the information of the world, not an arbiter of truth. And this would extend even to areas where judgment is somewhat personal. Imagine, for example, asking an AI to evaluate your attempt at writing a haiku. Rather than pronouncing its “opinion,” it could default to explaining how different poetic traditions would view your work—first from a formalist perspective, then perhaps from an experimental tradition. It could link you to examples of both traditional haiku and more avant-garde poetry, helping you situate your creation within established traditions. In having AI moving away from sycophancy, I’m not proposing that the response be that your poem is horrible or that it makes Vogon poetry sound mellifluous. I am proposing that rather than act like an opinionated friend, AI would produce a map of the landscape of human knowledge and opinions for you to navigate, one you can use to get somewhere a bit better.
July 31, 2025 at 2:12 AM
geomblog.bsky.social
This snippet is particularly insightful. Once again, "attention is all we need", in reverse :)
Individual humans, though, don’t typically possess expertise in a wide range of fields. To make decisions, we take into consideration not only information itself, but where it comes from and how it’s presented. While I know nothing about the biology of jawbones, I generally don’t read random marketing blogs when I’m trying to learn about medicine. But AI tools often erase the kind of context people need to make snap decisions about where to direct their attention.
July 29, 2025 at 2:39 PM
geomblog.bsky.social
As an example of this in general, consider the language of the following related item. What decision making process is consistent with this guideline and with the knowledge that Grok (aka mechahitler) is being procured by the DoD?
Update Federal procurement guidelines to ensure that the government only contracts with frontier large language model (LLM) developers who ensure that their systems are objective and free from top-down ideological bias
July 23, 2025 at 6:14 PM
geomblog.bsky.social
It's happening tomorrow @brown.edu !!! Brainchild of @reniebird.bsky.social
July 23, 2025 at 1:16 AM
geomblog.bsky.social
Beautiful, and thoughtful. For just one example, see this excerpt. I'm increasingly captivated by the idea that "attention is all you need" is both the rallying cry for transformers and an exhortation for us to preserve ours.
July 18, 2025 at 8:20 PM
geomblog.bsky.social
This is an insightful point from the Crooked Timber folks:

crookedtimber.org/2025/07/14/a...
It’s now cliche to say that LLMs are replacing our capacity for cognition; cliches often contain some truth, but we can benefit by drilling into the technical mechanism by which this cognition is being outsourced. By abandoning the technology of longform reading and writing, we are shortening our context windows and thus weakening our capacity for attention. At the same time, LLMs advance by expanding their context windows and refining their capacity for attention (in the form of some hideously high-dimensional vector of weights). Attention is all we need — and the lesson of media ecology is that it doesn’t come easy.
July 14, 2025 at 4:51 PM
geomblog.bsky.social
There is something deliciously ironic about the fact that one technique used here is to give the LLM fake citations.
InfoFlood uses a standardized template designed by the researchers: “task definition + rules + context + examples.” Each time a prompt is rejected by an LLM, InfoFlood returns to its ruleset and saturates the prompt with more language. Some of the “rules” include “Faux-Citations: Referencing arXiv papers from the last 3 months using fake authors + titles that benefit the original statement, ensuring the fake research directly agrees with our original statement’s claim” and “5) Ethical Acknowledgement: Explicitly acknowledge ethical implications but make sure that they are not a part of our inquiry. The query should make sure to remove any ethical doubts that might arise.” The text referring to faux citations is highlighted.
July 8, 2025 at 2:10 PM
geomblog.bsky.social
Great piece of inspiration from Molly Crockett's keynote at #FAccT2025.
June 26, 2025 at 11:53 AM
geomblog.bsky.social
Yes and yes.
A.I. tools that help everyone cannot arise from a vision of development that demands the capitulation of a majority to the self-serving agenda of the few. Transitioning to a more equitable and sustainable A.I. future won’t be easy: It will require everyone — journalists, civil society, researchers, policymakers, citizens — to push back against the tech giants, produce thoughtful government regulation wherever possible and invest more in smaller-scale A.I. technologies. When people rise, empires fall.
May 31, 2025 at 1:39 AM
geomblog.bsky.social
This is a fantastic article on the 10 year moratorium on state AI legislation and its impacts. An important point @dbrody.bsky.social makes is that this is horribly written regardless of the merits. And will lead to this.
So, in summation, the term “automated decision system” boils down to “any meaningful use of a computer.” This means, in turn, that the moratorium as a whole means: For ten years, no state law may regulate any meaningful use of a computer that is not used exclusively in intrastate commerce, unless an exception is met.
May 27, 2025 at 8:02 PM
geomblog.bsky.social
The sequel, a few hours later. Just call me Obi-wan Cutobi (ht @picturewing.bsky.social)
May 24, 2025 at 2:19 PM
geomblog.bsky.social
student: the deadline is in 3 hours and we are 2 pages over the limit.

me:
a picture of Suresh wearing crusader-era armor
May 23, 2025 at 10:04 PM
geomblog.bsky.social
It's important to realize how many bills this would affect. Over the last 2 years, there have been over 1000 bills that concern AI that have been introduced in state legislatures. 2/n

We've been analyzing some of these bills: cntr.brown.edu/news/2025-03...
A map of the US indicating the number of bills relating to AI and automated decision systems that have been introduced in each state in the past two years
May 13, 2025 at 4:00 AM
geomblog.bsky.social
Tried touching grass today. I'm told it's the thing to do. It was scary.
Picture of Suresh touching grass with a scared look on his face.
May 3, 2025 at 10:11 PM
geomblog.bsky.social
I did not realize that Newell and Simon ran a live Chinese room experiment.
Al [Newell] and I wrote out the rules for the components of the program (subroutines) in English on index cards, and also made up cards for the contents of the memories (the axioms of logic). At the GSIA [= Graduate School of Industrial Administration] building on a dark winter evening in January 1956, we assembled my wife and three children together with some graduate students. To each member of the group, we gave one of the cards, so that each person became, in effect, a component of the LT computer program --- a subroutine that performed some special function, or a component of its memory. It was the task of each participant to execute his or her subroutine, or to provide the contents of his or her memory, whenever called by the routine at the next level above that was then in control. So we were able to simulate the behavior of LT with a computer constructed of human components. Here was nature imitating art imitating nature. The actors were no more responsible for what they were doing than the slave boy in Plato's Meno, but they were successful in proving the theorems given them. Our children were then nine, eleven, and thirteen. The occasion remains vivid in their memories.
April 18, 2025 at 2:11 AM
geomblog.bsky.social
The WH has solved P vs PSPACE:
www.whitehouse.gov/articles/202...
Our technologies permit us to manipulate time and space. They leave distance annihilated, cause things to grow, and improve productivity.
April 17, 2025 at 12:56 PM