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.

You can literally just help people www.slowboring.com/p/give-yours...

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...

A classic

In case you think not being as good as your friends means you're not cut out to succeed.

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.

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

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...

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...

Single-blind!

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.

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.

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.

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.

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...

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.

Keep fighting the good fight, folks. It shouldn't be this hard.

I'm glad Australia is doing an experiment for us.

In the Messy Middle: Observations from the Front Line at the UKSG Forum - The Scholarly Kitchen

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/...

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.

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-...

Serious conversations that are representative do happen. They mostly don't happen in public online fora.

Christmas came early this year! Very happy to see our paper out in Science Advances. Led by @lfoswaldo.bsky.social, we ran a unique collective field-experiment on Reddit, to better understand who is participating in online debates and why. Paper: www.science.org/doi/10.1126/... And more below 👇

UK AISI is hiring for a technical research role on open-weight model safeguards. www.aisi.gov.uk/careers

Popularity is orthogonal to taste, and nowhere is that more true than with food: laurenleek.substack.com/p/how-google... The price of having a decent algorithmic social feed is constantly pressing "show less like this" on ragebait. The price of getting good food recommendations is ... ?

The big question: will "provably safe RL" in self-driving generalize? Are they solving the right problems with the right methods based on a accurate model of the world?

Bold strategy from India: Nationwide license to copyrighted content for AI training in exchange for payments into a national collections body. techcrunch.com/2025/12/09/i... The simplicity and straightforwardness of the approach is a bit refreshing, I gotta say. Will it work for creators?

I wanted dinner recommendations so I scraped 13,000+ London restaurants and accidentally discovered Google Maps is running a shadow economy. Anyway here's a dashboard and a political economy thesis: open.substack.com/pub/laurenle...

Cue the conspiracy theories that librarians are coordinating to suppress information (that actually doesn't exist).

Blogs deserve love too (and interoperability) ❤️ Rogue Scholar (rogue-scholar.org) makes science blogs more findable and citable: full-text search, long-term archiving, DOIs, metadata... You can register your blog and search ~200 blogs already there. Wrote more at www.linkedin.com/posts/alelaz...

Also, our team is hiring an AI Research Scientist! www.usajobs.gov/job/851528400

So, it's Monday, and there's no way to say this that doesn't sound stupid, so I'm just going to come out and say it: I've decided to Twitch stream the entire writing of an NIH R01 proposal, and you can follow along at twitch.tv/ColinDoesScience

I really wish people would stop saying "science", when they mean theorem proving or data analysis. There is currently no tool that can replace scientists, and I've worked extensively with essentially all the tools that claim to do so. You must think about process, not just outcomes!

Did you know that one base model is responsible for 94% of model-tagged NSFW AI videos on CivitAI? This new paper studies how a small number of models power the non-consensual AI video deepfake ecosystem and why their developers could have predicted and mitigated this.

Down with the Tyranny of the Mid!

Don't forget that just a few years ago, all the moderators of the biggest subreddits collectively protested Reddit forcing them to allow AI slop. Those who refused were relieved of admin duties for the communities they built.

The median academic opinion about AI has been 2017: quant methods are no use for qualitative problems 2021: these models are just parroting memorized phrases 2025: now that they do everything, higher ed is doomed So, if you're depressed right now, remember: we're almost certainly wrong! +

Two new recordings from our AGI Journalism Workshop with @tarbellcenter: @gleave.me: AI systems can fake alignment during training. Scalable oversight & lie detectors can reduce this. @alexbores.nyc: NY's RAISE Act passed with bipartisan support despite lobbying. 👇

US CAISI (the equivalent of the US "AI Safety Institute") just put out their approach to AI measurement & there's such a significant portion on construct validity (nist.gov/blogs/caisi-...). Great to see this after ongoing advocating about this issue (arxiv.org/abs/2511.04703)!

The enshittification commenceth.

"superintelligent AI could replace humans in controlling the planet" Bernie Sanders is right that this is a real risk that requires urgent attention. www.theguardian.com/commentisfre...

@richardsever.bsky.social @aarontay.bsky.social Might be worth mentioning this unfortunate default when speaking to people about using LLMs for literature retrieval. Haven't examined this systematically.

Asking LLMs for press releases and blogs to be excluded when asking for research summaries improves results. When I subsequently asked ChatGPT5 Pro for a Google Scholar search string, I got a good search string that also included -site:biorxiv.org -site:medrxiv.org -site:arxiv.org 🤨

FAR.AI is hiring a Head of Engineering (Research). Lead and scale our technical team as we work with frontier labs and governments to advance AI safety. If you want to drive research that shapes the trajectory of advanced AI systems, we want to meet you.👇

Asking informally: does anyone know someone who might be interested in a postdoc focused on understanding changes in memory representations driven by attention using EEG? ⚡️Thanks!

I'm at NeurIPS & hiring for our pretraining safety team at OpenAI! Email me if you want to chat about making safer base models!