William Gunn
@metasynthesis.net
Communications and community. More here: http://synthesis.williamgunn.org/about/
Talk to me: https://calendar.app.google/z4KR3xfXTc178es47
Talk to me: https://calendar.app.google/z4KR3xfXTc178es47
United's GenAI flight status updates are not quite fully baked. Flights do not just fly back and forth between two airports, as it's apparently assuming.
December 26, 2025 at 11:26 PM
United's GenAI flight status updates are not quite fully baked. Flights do not just fly back and forth between two airports, as it's apparently assuming.
Reposted by William Gunn
Give yourself the gift of doing things
Taking action and helping others is good and feels good.
www.slowboring.com
December 24, 2025 at 11:49 AM
Reposted by William Gunn
OSTP has issued a request for information, please share with them your ideas for how to improve government functions related to science funding and policy
www.federalregister.gov/documents/20...
www.federalregister.gov/documents/20...
Notice of Request for Information; Accelerating the American Scientific Enterprise
The Office of Science and Technology Policy (OSTP) requests input from all interested parties on Federal policy updates that aim to accelerate the American scientific enterprise, enable groundbreaking...
www.federalregister.gov
December 23, 2025 at 10:56 PM
OSTP has issued a request for information, please share with them your ideas for how to improve government functions related to science funding and policy
www.federalregister.gov/documents/20...
www.federalregister.gov/documents/20...
A classic
When thinking about the dynamics of posting on social platforms like Twitter or bluesky, I so often think of the great Paul Ford essay on the fundamental question of the web: “Why Wasn’t I Consulted?”
www.ftrain.com/wwic.html
www.ftrain.com/wwic.html
December 23, 2025 at 1:30 AM
A classic
In case you think not being as good as your friends means you're not cut out to succeed.
"Most top achievers (Nobel laureates and world-class musicians, athletes, chess players) demonstrated lower performance than many peers during their early years. Across the highest adult performance, peak performance is negatively correlated with early performance" www.science.org/doi/10.1126/...
Recent discoveries on the acquisition of the highest levels of human performance
Scientists have long debated the origins of exceptional human achievements. This literature review summarizes recent evidence from multiple domains on the acquisition of world-class performance. We re...
www.science.org
December 20, 2025 at 6:11 PM
In case you think not being as good as your friends means you're not cut out to succeed.
Reposted by William Gunn
Over the past month we've partnered with @blueroseorg to survey 5,660 Americans about their AI usage habits. Today we're releasing the results.
A key takeaway: A majority of Americans use AI on a weekly basis, with 35% using ChatGPT, 24% Gemini, and 13% Meta AI.
A key takeaway: A majority of Americans use AI on a weekly basis, with 35% using ChatGPT, 24% Gemini, and 13% Meta AI.
December 20, 2025 at 12:01 AM
Over the past month we've partnered with @blueroseorg to survey 5,660 Americans about their AI usage habits. Today we're releasing the results.
A key takeaway: A majority of Americans use AI on a weekly basis, with 35% using ChatGPT, 24% Gemini, and 13% Meta AI.
A key takeaway: A majority of Americans use AI on a weekly basis, with 35% using ChatGPT, 24% Gemini, and 13% Meta AI.
Reposted by William Gunn
Also once this ghost references makes it out to the web, LLMs that use the web to do RAG will be vulnerable. I tried chatgpt free version and it happily said a fake paper exists pointing to a real paper with the fake ref. The stronger paid model figured out and stated was probably a fake ref
December 20, 2025 at 9:34 AM
Also once this ghost references makes it out to the web, LLMs that use the web to do RAG will be vulnerable. I tried chatgpt free version and it happily said a fake paper exists pointing to a real paper with the fake ref. The stronger paid model figured out and stated was probably a fake ref
Reposted by William Gunn
Join our Replicability Project: Health Behavior!
We have 55 replication studies underway, our target is 65-70.
We are only recruiting for secondary data replications--i.e., using existing data to test the original question.
Here's a list of studies we think could be feasible.
If interested...
We have 55 replication studies underway, our target is 65-70.
We are only recruiting for secondary data replications--i.e., using existing data to test the original question.
Here's a list of studies we think could be feasible.
If interested...
Replications Sourcing Sheet
docs.google.com
December 18, 2025 at 2:29 PM
Join our Replicability Project: Health Behavior!
We have 55 replication studies underway, our target is 65-70.
We are only recruiting for secondary data replications--i.e., using existing data to test the original question.
Here's a list of studies we think could be feasible.
If interested...
We have 55 replication studies underway, our target is 65-70.
We are only recruiting for secondary data replications--i.e., using existing data to test the original question.
Here's a list of studies we think could be feasible.
If interested...
Reposted by William Gunn
Are you passionate about exploring what a conceptually cogent, methodologically sound, and well-founded AI evaluation and safety research might look like? Come do a PhD with us.
Closing Date: 10 February 2026
Apply here aial.ie/hiring/phd-a...
Closing Date: 10 February 2026
Apply here aial.ie/hiring/phd-a...
December 17, 2025 at 6:52 PM
Are you passionate about exploring what a conceptually cogent, methodologically sound, and well-founded AI evaluation and safety research might look like? Come do a PhD with us.
Closing Date: 10 February 2026
Apply here aial.ie/hiring/phd-a...
Closing Date: 10 February 2026
Apply here aial.ie/hiring/phd-a...
If you optimize for engagement and there's nobody empowered & willing to say, "We'll take the engagement hit & not launch this", you're going to get sycophancy. It's going to hit some people really hard and then you'll get hit hard with worse regulation than if you had showed any restraint or taste.
If we are banning cell phones for kids we need to be talking about banning chatbots for boomers
www.persuasion.community/p/my-chatgpt...
www.persuasion.community/p/my-chatgpt...
December 17, 2025 at 7:44 PM
If you optimize for engagement and there's nobody empowered & willing to say, "We'll take the engagement hit & not launch this", you're going to get sycophancy. It's going to hit some people really hard and then you'll get hit hard with worse regulation than if you had showed any restraint or taste.
Whenever you see an article that prints a quote from a spokesperson and then says something like "to better understand, we spoke to 18 current and former employees", you know they're about to throw down.
December 17, 2025 at 7:39 PM
Whenever you see an article that prints a quote from a spokesperson and then says something like "to better understand, we spoke to 18 current and former employees", you know they're about to throw down.
Browser makers, particularly Chrome and Safari, have long wanted to remove the URL bar and disallow manual typing of URLs. They want the browser to become TV. I don't care if 99.99% of users never type a URL. They'll have to pry it from my cold dead hands.
December 16, 2025 at 5:35 PM
Browser makers, particularly Chrome and Safari, have long wanted to remove the URL bar and disallow manual typing of URLs. They want the browser to become TV. I don't care if 99.99% of users never type a URL. They'll have to pry it from my cold dead hands.
Reposted by William Gunn
AI-assisted coding: 10 simple rules to maintain scientific rigor www.thetransmitter.org/artificial-i... - my latest in @thetransmitter.bsky.social
AI-assisted coding: 10 simple rules to maintain scientific rigor
These guidelines can help researchers ensure the integrity of their work while accelerating progress on important scientific questions.
www.thetransmitter.org
December 16, 2025 at 1:48 PM
AI-assisted coding: 10 simple rules to maintain scientific rigor www.thetransmitter.org/artificial-i... - my latest in @thetransmitter.bsky.social
This feed has totally replaced the Discover feed for me. It seems to take popularity less into account and "show less like this" actually works.
Enjoying the For You feed? Give it a like ♡ to help more people discover it: bsky.app/profile/did:...
The more people use it -> the more feedback we get -> the better we can make it for you.
The more people use it -> the more feedback we get -> the better we can make it for you.
December 16, 2025 at 5:19 PM
This feed has totally replaced the Discover feed for me. It seems to take popularity less into account and "show less like this" actually works.
Reposted by William Gunn
Nice write-up of the first wave of the AI Safety Red Team Clinic at @cornelltech.bsky.social.
We help nonprofits & public sector orgs stress-test their AI tools before they go live thru a *free* 6-week adversarial testing exercise.
We're looking for our next client! Please help spread the word.
We help nonprofits & public sector orgs stress-test their AI tools before they go live thru a *free* 6-week adversarial testing exercise.
We're looking for our next client! Please help spread the word.
‘Red team’ students stress-test NYC health department’s AI | Cornell Chronicle
People usually strive to be their true, authentic selves, but this fall, five master’s students at Cornell Tech adopted not only alter egos but also “bad intent,” in an effort to make AI safer for hea...
news.cornell.edu
December 16, 2025 at 4:36 PM
Nice write-up of the first wave of the AI Safety Red Team Clinic at @cornelltech.bsky.social.
We help nonprofits & public sector orgs stress-test their AI tools before they go live thru a *free* 6-week adversarial testing exercise.
We're looking for our next client! Please help spread the word.
We help nonprofits & public sector orgs stress-test their AI tools before they go live thru a *free* 6-week adversarial testing exercise.
We're looking for our next client! Please help spread the word.
I've always thought these systems are bizarre. I know people want to be able to analyze search performance in ways they're accustomed to, but it's like going to all the trouble to make a car, then taking out the engine and bolting the frame to a horse. If you already have an index, I get it, but...
Insight: I've been quite negative about those simple "push natural language input to LLM to generate Boolean & run" systems than many librarians mostly because I have higher expectations than them.. (1)
December 16, 2025 at 5:16 PM
I've always thought these systems are bizarre. I know people want to be able to analyze search performance in ways they're accustomed to, but it's like going to all the trouble to make a car, then taking out the engine and bolting the frame to a horse. If you already have an index, I get it, but...
Reposted by William Gunn
Celebrating Public-Good Curators: An Interview with Tracey Brown and Camille Gamboa scholarlykitchen.sspnet.org/2025/12/16/c...
Celebrating Public-Good Curators: An Interview with Tracey Brown and Camille Gamboa - The Scholarly Kitchen
Who are public-good curators and how can they help improve public trust in science? Learn more in this interview with Tracey Brown (Sense about Science) and Camille Gamboa (Sage) about their recently ...
scholarlykitchen.sspnet.org
December 16, 2025 at 10:58 AM
Celebrating Public-Good Curators: An Interview with Tracey Brown and Camille Gamboa scholarlykitchen.sspnet.org/2025/12/16/c...
I wouldn't go as far as the author and say measurement itself is the problem (historians have been particularly ill-served by the institutionalization of citations) but the increased rate at which LLMs can create the structure but not the substance of a scientific publication is concerning.
Absolutely essential piece by @kevinbaker.bsky.social critiquing AI automation of science by pulling back the curtain on what modern science actually is. LLMs already replicate and accelerate its irrational bureaucratic structure.
Context Widows
or, of GPUs, LPUs, and Goal Displacement
artificialbureaucracy.substack.com
December 15, 2025 at 7:38 PM
I wouldn't go as far as the author and say measurement itself is the problem (historians have been particularly ill-served by the institutionalization of citations) but the increased rate at which LLMs can create the structure but not the substance of a scientific publication is concerning.
Keep fighting the good fight, folks. It shouldn't be this hard.
You might be surprised how much more fighting that approach entails...
www.sfchronicle.com/bayarea/arti...
www.sfchronicle.com/bayarea/arti...
Silicon Valley school district agrees to regrade science exam that sparked firestorm
After a monthslong internal investigation, irate emails and an international outcry, a Silicon Valley school district has confirmed that burning firewood does not, in fact, produce oxygen.
www.sfchronicle.com
December 15, 2025 at 6:35 PM
Keep fighting the good fight, folks. It shouldn't be this hard.
I'm glad Australia is doing an experiment for us.
A new PNAS paper finds that polarization increased immediately after the invention of smartphones and the advent of social media, which both appeared around the same year, 2008.
www.pnas.org/doi/10.1073/...
www.pnas.org/doi/10.1073/...
December 15, 2025 at 6:24 PM
I'm glad Australia is doing an experiment for us.
Reposted by William Gunn
In the Messy Middle: Observations from the Front Line at the UKSG Forum - The Scholarly Kitchen
In the Messy Middle: Observations from the Front Line at the UKSG Forum - The Scholarly Kitchen
The UKSG Forum is "an entire 2-3 day conference stripped back to bare essentials and completed in just one day". Here are the key takeaways.
scholarlykitchen.sspnet.org
December 15, 2025 at 12:30 PM
In the Messy Middle: Observations from the Front Line at the UKSG Forum - The Scholarly Kitchen
Reposted by William Gunn
We're hiring interns in the Computational Social Science group at Microsoft Research NYC!
If you're interested in designing AI‑based systems and understanding their impact at both individual and societal scales, apply here by Jan 9, 2026: apply.careers.microsoft.com/careers/job/...
If you're interested in designing AI‑based systems and understanding their impact at both individual and societal scales, apply here by Jan 9, 2026: apply.careers.microsoft.com/careers/job/...
Research Intern - Computational Social Science | Microsoft Careers
Research Interns put inquiry and theory into practice. Alongside fellow doctoral candidates and some of the world's best researchers, Research Interns learn, collaborate, and network for life. Researc...
apply.careers.microsoft.com
December 15, 2025 at 4:33 PM
We're hiring interns in the Computational Social Science group at Microsoft Research NYC!
If you're interested in designing AI‑based systems and understanding their impact at both individual and societal scales, apply here by Jan 9, 2026: apply.careers.microsoft.com/careers/job/...
If you're interested in designing AI‑based systems and understanding their impact at both individual and societal scales, apply here by Jan 9, 2026: apply.careers.microsoft.com/careers/job/...
We can get the benefits while minimizing the harms if just a key participants would stop enabling it. If the AI industry can't get @hf.co & CivitAI to do literally anything at all to stop deepfakes, they're going to bring far more onerous regulation down on everyone.
🧵🧵🧵 In the past few months, I have looked at hundreds, maybe thousands, of AI porn images/videos (for science).
Here's what I learned from our investigation of over 50 platforms, sites, apps, Discords, etc., while writing this paper.
papers.ssrn.com/sol3/papers...
Here's what I learned from our investigation of over 50 platforms, sites, apps, Discords, etc., while writing this paper.
papers.ssrn.com/sol3/papers...
December 15, 2025 at 6:14 PM
We can get the benefits while minimizing the harms if just a key participants would stop enabling it. If the AI industry can't get @hf.co & CivitAI to do literally anything at all to stop deepfakes, they're going to bring far more onerous regulation down on everyone.
Reposted by William Gunn
Download our large database of postdoc fellowships in all fields of research.
Database freely available to all; 281 fellowships.
Download here: research.jhu.edu/rdt/funding-...
Database freely available to all; 281 fellowships.
Download here: research.jhu.edu/rdt/funding-...
December 13, 2025 at 1:38 PM
Download our large database of postdoc fellowships in all fields of research.
Database freely available to all; 281 fellowships.
Download here: research.jhu.edu/rdt/funding-...
Database freely available to all; 281 fellowships.
Download here: research.jhu.edu/rdt/funding-...