Morteza Dehghani

Our new study in Psych Science @psychscience.bsky.social reveals many interesting-- and concerning -- things about the language on kids' shows, including a grammatical bias in favor of men and boys as agents or "doers" 👇

Syntactic and Semantic Gender Biases in the Language on Children’s Television: Evidence From a Corpus of 98 Shows From 1960 to 2018 - Andrea C. Vial, Aida Mostafazadeh Davani, Ruyuan Zuo, Shreya Haval... Biased media content shapes children’s social concepts and identities. We examined gender bias in a large corpus of scripts from 98 children’s television progra...

🔥Excited to share our new paper in PsychScience! Shout out to best-ever collaborators Aida Davani, @ruyuan-zuo.bsky.social, @shreyahavaldar.bsky.social, Eleanor Chestnut, @mdehghani.bsky.social & @andreicimpian.bsky.social 🧵1/8…
doi.org/10.1177/0956...

Understanding this dynamic is crucial. External calls for regime change often backfire, strengthening the regime's narrative. The path forward is a psychological shift, not a political one. (2/2)

Opinion | The Moral Paralysis Facing Iranians Right Now

My new op-ed in the @nytimes.

I argue that Iranians are caught in a state of “moral paralysis,” a psychological trap set by the Islamic Republic itself. It pits two of our most sacred values against each other: liberation vs. self-determination. (1/2) www.nytimes.com/2025/06/28/o...

Check up on your Iranian colleagues and students; this is a full-on war!

Watch this montage of clips from Israel's Channel 14, filled with racist and genocidal bile, shocking stuff, ask yourself why you don't see these clips on US or UK mainstream news, and then read Prem's deep dive for Zeteo into Channel 14:
zeteo.com/p/inside-the...

PhD preferred.

This position is remote and can be anywhere in the U.S. If interested, send me a resume. I won’t be able to respond to all messages.

This is a unique opportunity to contribute fundamental research, while also applying and operationalizing your findings to build cutting-edge AI evaluation systems in a fast-paced, impactful environment!

We’re looking for a researcher to work on methods and building tools for AI evaluation. We are seeking candidates with expertise at the intersection of agentic systems and personalization, demonstrated by a strong publication record in relevant venues.

Gaza: Latest Israeli Plan Inches Closer to Extermination The Israeli government’s plan to demolish what remains of Gaza’s civilian infrastructure and concentrate the Palestinian population into a tiny area would amount to an abhorrent escalation of its ongo...

This is what Leibowitz was warning about when he referred to Judeo Nazis www.hrw.org/news/2025/05...

I am very sorry to hear that, Maddy.

Video Shows Aid Workers Killed in Gaza Under Gunfire Barrage, With Ambulance Lights On

“There is no God but God, Muhammad is his messenger,” the paramedic is heard saying. He asks God for forgiveness and says he knows he is going to die.

“Forgive me, mother. This is the path I chose — to help people,” he said. “Allahu akbar,” God is great, he says.

www.nytimes.com/2025/04/04/w...

Tweet from Yuval Abraham: "A group of settlers just lynched Hamdan Ballal, co director of our film no other land. They beat him and he has injuries in his head and stomach, bleeding. Soldiers invaded the ambulance he called, and took him. No sign of him since."

BREAKING: Yuval Abraham, co-director of Oscar-winning documentary "No Other Land," says Israeli settlers beat his co-director Hamdan Ballal, injuring his head and stomach — and then Israeli soldiers invaded the ambulance he was in and seized him.

Ballal's whereabouts are now unknown.

Such a great read! Thanks

Anyway to share a version outside the paywall?

Very similar to the Islamic Republic

So Cal Stand Up for Science: Fri Mar 7 noon @the Wilshire Federal Building

The paper covers key considerations for using LLMs, including transparency, validation, and dealing with model updates. We also provide a checklist for ensuring high-quality research. This work was led by @sabdurah.bsky.social and @alirezaziabari.bsky.social

At @spspnews.bsky.social, many asked for guidelines on using LLMs in the social sciences. Our new paper in AMPPS addresses exactly that! We provide a primer for evaluating LLMs in social science research, focusing on methodological rigor, replicability, and validity: osf.io/preprints/ps...

Our new paper in PNAS Nexus. County morals > politics for carbon footprints? 🤯 We find that county-level values predict actual green behavior (emissions!), not just attitudes. Stronger than political leaning.

Very cool work on the intersection of interpretability and multi-lingual LLMs, led by @elnaz-rahmati.bsky.social

This project was led by @alirezaziabari.bsky.social bari.bsky.social and Nona Ghazizadeh, with collaborations with @zhpinkman.bsky.social ky.social, Farzan Karimi-Malekabadi, and @payampiray.bsky.social

Key Findings:

Accuracy-efficiency trade-off: System 2-aligned models excelled in arithmetic and symbolic reasoning, while System 1 models were better at commonsense reasoning.

Uncertainty and confidence: System 2 models showed greater uncertainty, while System 1 models were more confident.

Our new paper explores how to align LLMs with System 1 (intuitive) and System 2 (analytical) thinking styles. This work challenges the idea that step-by-step reasoning (CoT) is always best and highlights the need for adapting reasoning strategies based on the task
arxiv.org/abs/2502.12470

Key Findings:

Accuracy-efficiency trade-off: System 2-aligned models excelled in arithmetic and symbolic reasoning, while System 1 models were better at commonsense reasoning.

Uncertainty and confidence: System 2 models showed greater uncertainty, while System 1 models were more confident.

Our new paper explores how to align LLMs with System 1 (intuitive) and System 2 (analytical) thinking styles. This work challenges the idea that step-by-step reasoning (CoT) is always best and highlights the need for adapting reasoning strategies based on the task
arxiv.org/abs/2502.12470

This project was led by @alirezaziabari.bsky.social abari.bsky.social and Nona Ghazizadeh, with collaborations with @zhpinkman.bsky.social kman.bsky.social, Farzan Karimi-Malekabadi, and
@payampiray.bsky.social

Key Findings:

Accuracy-efficiency trade-off: System 2-aligned models excelled in arithmetic and symbolic reasoning, while System 1 models were better at commonsense reasoning.

Uncertainty and confidence: System 2 models showed greater uncertainty, while System 1 models were more confident.

This project was led by @alirezaziabari.bsky.social and Nona Ghazizadeh, with collaborations with @zhpinkman.bsky.social, Farzan Karimi-Malekabadi, and @payampiray.bsky.social