Rafael M Batista
LightNews LightNews
rafmbatista.bsky.social
Rafael M Batista
@rafmbatista.bsky.social
4.1K followers 820 following 610 posts
Behavioral Scientist.

Lately, I've been thinking (and posting) about: AI+Psych, Personal Finance, Consumer Behavior. Civically engaged, so I occasionally post about that too.
Posts Replies Media Videos
rafmbatista.bsky.social
Make America Make Sense Again
November 29, 2025 at 9:31 PM
rafmbatista.bsky.social
these images were generated with google's latest nana banana pro tool. we have no idea how bad it is about to get.
Actual image of me leaving an ice-cream shop in Denver. Taken by Yena on an iPhone. AI-generated from first image with prompt “Recreate this image out of ice cream” Another AI-generated image meant to look like the ice-cream diorama was melting. Three attempts and it’s still shite if you ask me. AI-generated image from the previous with prompt “Now zoom out. Show the photograph of the set where the diorama is being shot. The diorama should be small but clearly containing same features as this image. The photo should be one of the studio “behind the scenes” where we see the crew setting up this shot and the camera where this shot was taken” Camera angle is not the one from the photo and not the one that took the photo in the screen. No hand in photo. Impressive? Sure. But still clearly AI slop.
November 27, 2025 at 12:31 PM
rafmbatista.bsky.social
This week I was in Florence, Italy participating in a #BehSci + #AI conference hosted by @unicef.org & Gates Foundation

Really well organized w/ nice mix of academics + practitioners, beh scientists + data scientists
November 21, 2025 at 11:00 PM
rafmbatista.bsky.social
lol. Dude hasn’t seen the emails 😂

He’s brazen. Or maybe just detached from reality. #EpsteinEmails
November 14, 2025 at 6:30 PM
rafmbatista.bsky.social
What do we want from thought partners? 🤔

This work proposes 3 things:
1️⃣ You understand me
2️⃣ I understand you
3️⃣ We understand the world

Awesome research I learned of this week, led by Katie Collins w/ several of my colleagues here at @princeton.edu + @cocoscilab.bsky.social

doi.org/10.1038/s415...
A text excerpt titled “Desiderata.” It discusses what people want from thought partners and outlines three desiderata for human-centered thought partners. 1. You understand me: Thought partners should understand our goals, plans, beliefs, and resource limits to collaborate effectively, adapting strategies based on past and present observations. 2. I understand you: Thought partners should act and communicate in ways that are legible and intuitive to humans. 3. We understand the world: Thought partners should be accurate and knowledgeable, sharing a representation of the world or task. The text emphasizes synergy—thought partnerships being more than the sum of their parts.
November 12, 2025 at 12:59 PM
rafmbatista.bsky.social
If you’re going to have a sham election, at least make up better numbers

Does anyone really believe a single individual is supported by 98% of a voting population?

I don’t even believe Taylor Swift would win prom queen with those number
Reporting by The Economist magazine: “Samia Suluhu Hassan, the incumbent, was declared the winner of Tanzania’s presidential election. The electoral commission, whose members are picked by the president, said Mrs Samia was re-elected with 98% vote. The commission barred the main opposition from standing. Hundreds of people are reported dead following days of anti-government violence. An internet blackout and curfew are in place.”
November 1, 2025 at 12:33 PM
rafmbatista.bsky.social
This week @mdubova.bsky.social gave a talk at Princeton

It was an excellent talk! Marina studies scientific discovery–how scientific discoveries shape concepts AND how the mental concepts shape discovery

She shared a ton of empirical work she's done and her theories were so interesting
Screenshot. Text reads: "The title and abstract for the talk are: Cognitive Mechanisms of Discovery To understand and navigate our world, both individuals and scientific communities create simplifying representations, such as concepts and theories. How do we construct useful representations from our experiences, and how do we use these representations to guide our learning? In this talk, I discuss empirical research on the mechanisms of human concept learning, highlighting the ease with which we adopt new, even arbitrary, conceptualizations. I illustrate how these acquired concepts shape how we perceive and explore the world. For example, I show that our perception of objects becomes biased by our conceptual needs and our knowledge about these objects. I suggest that similar mechanisms are at play when scientific conceptualizations, such as the DSM in psychopathology and the periodic table in chemistry, guide scientific exploration. Then, I discuss the double-edged nature of theory-guided discovery—although conceptualizations can efficiently steer us towards new experiences that further refine our knowledge, they can also lead our exploration astray. I present a computational model of scientific discovery in which agents conduct experiments, build theories, and share results to advance collective understanding of the world. The model reveals that when new experiments are guided by existing theoretical frameworks, scientific communities risk missing important aspects of the world not yet captured by their theories. I conclude by reviewing our current and future research that integrates cognitive psychology, machine learning, and philosophy of science to enhance our understanding of how theories and observations can inform each other to support—rather than hinder—human learning and scientific progress."
October 31, 2025 at 1:38 PM
rafmbatista.bsky.social
#Vote 🗳️
Photo “I Voted Early” sticker being held in someone’s hand
October 25, 2025 at 7:39 PM
rafmbatista.bsky.social
I’m picturing Ortaköy in Istanbul
October 25, 2025 at 4:59 PM
rafmbatista.bsky.social
U.S. Dept of Defense announced today 26 journalists signed their agreement.

These journalists are not serious. None of the major news orgs signed on. Can’t trust what they write if they’ve agreed to have it vetted by DoD. #propaganda
October 23, 2025 at 12:57 AM
rafmbatista.bsky.social
October 22, 2025 at 11:10 PM
rafmbatista.bsky.social
Wait, but have you tried the Genmoji’s??
October 19, 2025 at 9:28 PM
rafmbatista.bsky.social
#NoKings in Princeton, NJ
October 18, 2025 at 8:35 PM
rafmbatista.bsky.social
I also discovered one! New species of barking gecko 🦎
October 2, 2025 at 3:18 AM
rafmbatista.bsky.social
The disclaimer links to the following excerpt

support.claude.com/en/articles/...
Screenshot that reads: "Security Considerations The create files feature gives Claude a sandboxed computing environment with limited internet access. Internet access lets Claude download and use packages (e.g., npm). It is possible for a bad actor to inconspicuously add instructions via external files or websites that trick Claude into: Downloading and running untrusted code in the sandbox environment for malicious purposes. Reading sensitive data from a Claude connected knowledge source (e.g., Remote MCP, projects) and using the sandbox environment to make an external network request to leak the data. This means Claude can be tricked into sending information from its context (e.g., prompts, projects, data via MCP, Google integrations) to malicious third parties. To mitigate these risks, we recommend you monitor Claude while using the feature and stop it if you see it using or accessing data unexpectedly. You can report issues to us using the thumbs down function directly in claude.ai. In line with our safe and trustworthy agents framework, we have applied the following mitigations for individual users: " Screenshot that reads: "In line with our safe and trustworthy agents framework, we have applied the following mitigations for individual users: Given you full control of the feature. You can turn it on or off at any time. Designed Claude to give you user-friendly summaries of its actions so you can see what it is doing. You can stop Claude’s actions at any time and we recommend monitoring Claude’s work while using the feature. Given you the ability to review and audit actions taken by Claude within the Sandbox environment. Disabled public sharing of conversations that include any file artifacts from the create files feature for Pro and Max users. Limited the duration of tasks that can be completed by Claude and the length of time you can use a single sandbox container to avoid loops of malicious activity. Implemented sandbox isolation such that no sandbox environments are ever shared between users. Intentionally limited the network, container, and storage resources. Implemented a classifier to detect prompt injections and stop execution if they are detected. We have performed red-teaming and security testing on the feature. We have a continuous process for ongoing security testing and red-teaming of this feature. We encourage organizations to evaluate these protections against their specific security requirements when deciding whether to enable this feature."
September 30, 2025 at 10:40 PM
rafmbatista.bsky.social
Screenshot of New York Times article headline that reads: “A Novice Defense Secretary Lectures the Brass on What It Takes to Win Senior officers, summoned from around the world, are entrusted to manage complex military operations. They got a lecture on fitness and grooming standards.” Source: https://www.nytimes.com/2025/09/30/us/politics/hegseth-military-officers.html?smid=nytcore-ios-share&referringSource=articleShare
September 30, 2025 at 9:24 PM
rafmbatista.bsky.social
Noticing that @economist.com is using POTUS’s term, “secretary of war” while @nytimes.com refers to “Defense Secretary. NYT go so far as to explain why, which I think is better reporting
Screenshot from the New York Times that reads: “Gen. Dan Caine, the chairman of the Joint Chiefs of Staff, introduced Defense Secretary Pete Hegseth, calling him the Secretary of War, before Hegseth took to the podium. President Trump has renamed the department but Congress has not yet approved the change.” Source: https://www.nytimes.com/live/2025/09/30/us/trump-government-shutdown-hegseth/4a3b421b-60a9-587e-b80b-690752073224?smid=url-share
September 30, 2025 at 12:56 PM
rafmbatista.bsky.social
#AI system designed by Palantir to facilitate immigration enforcement by #ICE (reported in @economist.com)

I can’t imagine this is going to go well

I don’t invest in Palantir because of stuff like things l
Screenshot of Economist article that reads: “America’s Immigrations and Customs Enforcement is having trouble meeting its targets. In June it arrested 30,000 people. That is a lot compared with the 8,000 arrested in that month the year before. But it is far short of the administration’s target of 90,000 a month. Like many unimaginative employers, the administration has decided that the solution is artificial intelligence. This week it began using ImmigrationOS, an AI system designed by Palantir, a data firm named after the all-seeing stones in J.R.R. Tolkien’s “The Lord of the Rings”. The system will sift through huge amounts of data—some of it collected with questionable legal authority from government agencies—to help ICE with “streamlining selection and apprehension” of illegal immigrants. Given the well-documented problems and biases of AI systems, critics worry that ImmigrationOS might lead to wrongful arrests and deportations. But considering the administration’s love of security theatre, this may be a feature, not a bug. “
September 27, 2025 at 4:29 PM
rafmbatista.bsky.social
Here is a poster I presented today at the Human-AI Complementarity workshop hosted by the NSF #AI Institute for Societal Decision Making

You can read the latest version of the paper here: www.rafaelmbatista.com/jmp/

This is joint work with the wonderful @jamesross0.bsky.social
An academic poster titled "Words that work: Using large language models to generate and refine hypotheses" by Rafael M. Batista & James Ross. The poster presents a framework that combines Large Language Models (LLMs), Machine Learning (ML), and human evaluation to generate interpretable hypotheses from text data. A central diagram shows a three-step process: 1. Hypothesize (using an LLM to generate a hypothesis from text pairs), 2. Intervene (rewriting text to incorporate the hypothesized feature), and 3. Predict (using an ML model to measure the effect of the change). The framework was applied to a dataset of news headlines to find what drives engagement. The poster lists generated hypotheses, such as "Framing a message with an element of surprise" or "Incorporating the concept of parody" increases engagement. A section on out-of-sample testing shows a regression table with statistically significant results for several hypotheses, including Surprise, Parody/Sarcasm, and Physical Reaction. A bar chart on the left illustrates the additional explanatory power gained from the newly discovered features compared to previously known ones. A QR code is in the top right corner. This alt-text was written by AI.
September 25, 2025 at 7:36 PM
rafmbatista.bsky.social
SCOTUS Justice Kagan makes an important critique of the Court’s use of emergency powers.

If Congress were made up of leaders these days they would push back to protect their independence.
Excerpt from a New York Times article that reads: “She also laid out a broader criticism of the court’s emergency action in this case and others, writing that the court’s emergency docket “should never be used, as it has been this year, to permit what our own precedent bars.” She added that “it should not be used, as it also has been, to transfer government authority from Congress to the president, and thus to reshape the nation’s separation of powers.” Source: https://www.nytimes.com/2025/09/22/us/politics/supreme-court-ftc-commissioner-firing.html?smid=nytcore-ios-share&referringSource=articleShare
September 22, 2025 at 11:25 PM
rafmbatista.bsky.social
They should’ve eliminated the tax cuts for the rich

It swamps the other provisions in the bill and makes it clear that in fact they’re taking away from most Americans to fund a handout for the rich

It’s a govt handout. That’s what this is
July 1, 2025 at 6:17 PM
rafmbatista.bsky.social
Final days on campus after 6+ years #ChicagoBooth
Wide-angle view of a modern, spacious atrium with high glass ceilings and large windows letting in natural light. The area is furnished with a mix of armchairs and tables arranged in open seating. A banner reading “Chicago Booth” is visible on the second floor, indicating this is part of the University of Chicago Booth School of Business. The architecture features clean lines, metal beams, and polished flooring, with plants placed around the seating area for decoration.
June 29, 2025 at 4:08 PM
rafmbatista.bsky.social
You’re so right! 😂😜
June 15, 2025 at 1:52 PM
rafmbatista.bsky.social
The article opens with this beautiful description of the #NoKings protest in #Chicago
Screenshot of the first two paragraphs of the article linked in the previous post, which reads: At MIDDAY, BY the time the “No Kings” march in Chicago was due to start, so many people were squashed into Daley Plaza, in the city’s downtown, that it caused a sort of gridlock, with protesters unable to march in any direction. There were a few stirring speeches—for those close enough to hear, anyway—but mostly people milled around enjoying the show. Some came in fancy dress: furries; Mexican wrestlers; clowns and two men dressed as cardboard tanks. Some carried megaphones. The communists, with hammer and sickle flags, were well prepared; the families with small children on cargo-bicycles less so. What united them all was disdain for Donald Trump. Eventually the protesters were able to march, and they spread across almost the entire loop, Chicago’s central business district. The police, in short sleeves and without riot gear, stood to the side. Amplifiers blared Sister Sledge and Rage Against the Machine. Men and women led chants of: “Hey ho, hey ho, Donald Trump has got to go” and “No Hate, No Fear, Immigrants are Welcome Here”. By 2pm or so, most protesters had found their way across the bridges and past Trump Tower, the president’s grandiose skyscraper hotel, where, despite obvious police nerves, nobody tried to do any damage. And then they dispersed, leaving the streets for bars, restaurants and El trains.
June 15, 2025 at 1:32 PM
rafmbatista.bsky.social
Thank you to the organizers and all the volunteers who made this possible. There were thousands of people out today in Chicago–this kind of participation is a sign of a still strong democracy ✊ #NoKings #Chicago #FlagDay #Resist
A person holds up a cardboard protest sign that covers their face. The sign features a cartoon depiction of Donald Trump as a baby sitting in a high chair, wearing a crown, and eating fast food. Fries and chicken nuggets are scattered on the tray, with a drink cup labeled “‘Merica.” The handwritten text at the top reads “MAN (baby) IN THE HIGH CHAIR.” In the background, several people walk past a granite building, including uniformed police officers and another protester wearing black. A large inflatable balloon shaped like a cartoonish baby version of Donald Trump is displayed during a protest in downtown Chicago. The balloon has an exaggerated expression with an open mouth and yellow hair. A man stands in front of the balloon holding a bike that is attached to the base of the balloon. Around them, people walk and hold signs. Visible flags include the American and Mexican flags. The background features tall buildings. A large group of protesters walks under a steel elevated train track in Chicago, holding signs above their heads. The signs include messages such as “NO KINGS IN AMERICA SINCE 1776,” “It shouldn’t have to happen to you for it to matter,” and “From PALESTINE to MEXICO all the walls have got to go!” The crowd includes people wearing hats, backpacks, and casual clothing. Some protesters carry flags, including a Palestinian flag. The urban street is filled with people stretching into the distance. A large group of protesters gathers outside Trump Tower in downtown Chicago. The reflective glass skyscraper dominates the background, with the word “TRUMP” in large silver letters across the middle of the building. Protesters fill the area in front of the tower, standing close together and holding signs, though individual signs are mostly too distant or obscured to read clearly. The sky is clear and blue, and neighboring high-rise buildings are visible on either side of the frame.
June 15, 2025 at 1:01 AM