←
Aida Davani9,252 posts
Aida Davani avatar

Aida Davani

@aidaa

Catsplaining science | CS PhD @USC | Research Scientist @Google | NLP/AI Ethics | She/her

◷ Joined September 2008▧ 90 Photos and videos
aleskxyz avatar
⟳ Aida Davani retweeted
aleskxyz@aleskxyz·Mar 11

از چند نفر شنیدم که این روش در کمال تعجب جواب داده! اپ mahsang رو توی حالت proxy اجرا میکنید و به یکی از کانفیگهای رایگان متصل میشید (مهم نیست وصل نمیشه، فقط بزنید فعال بشه) بعد توی سایفون اپ mahsang رو از vpn استثنا میکنید و توی کانفیگ پروکسی سایفون، اون رو به mahsang وصل میکنید. Host: 127.0.0.1 Port: 10809 بعد سایفون رو اجرا میکنید و میذارید تلاشش رو بکنه تا وصل بشه، قطعا زمان میبره ولی اگه متصل بشه سرعت خوبی میده.

1:41
Sasan Amjadi avatar
⟳ Aida Davani retweeted
Sasan Amjadi@SasanAmjadi·Mar 9

پلیس اعلام کرده است که مسعود مسجودی، ریاضیدان ایرانی و مدرس پیشین دانشگاه Simon Fraser که ماه گذشته در شهر برنابی ناپدید شده بود، به احتمال زیاد توسط عوامل حکومت ایران به قتل رسیده است. او یکی از منتقدان جدی باند پهلوی بود.

#REPORT: Police say that Masood Masjoody, an Iranian mathematician and former instructor at Simon Fraser University who went missing in Burnaby last month, was likely murdered by Iranian government agents.

Post attachment
Aida Davani avatar
Aida Davani@aidaa·Jan 21

Standardized AI governance may be insufficient. Categorizing human-like design as "inherently hazardous" ignores evidence that it builds necessary rapport in some cultures while inducing backlash in others.

Aida Davani avatar
Aida Davani@aidaa·Jan 21

To ensure robust measurement, we used an incentivized Trust Game with tangible stakes, rather than just self-reports. The evidence suggests anthropomorphic cues do not necessarily bypass deliberative cognitive processing to influence how users allocate resources.

Aida Davani avatar
Aida Davani@aidaa·Jan 21

Our results show divergent outcomes moderated by culture. 🇧🇷 In Brazil: High human-likeness increased engagement and social signals. 🇯🇵 In Japan: The exact same design levers resulted in decreased trust.

Aida Davani avatar
Aida Davani@aidaa·Jan 21

We experimentally manipulated AI human-likeness via design and sociality variables. Result: While we successfully increased anthropomorphism globally, this did NOT lead to a universal increase in trust. This contradicts the idea that human-likeness inherently causes over-trust.

Aida Davani avatar
Aida Davani@aidaa·Jan 21

What makes AI feel "human"? Users prioritize pragmatic, interactional cues, e.g., conversation flow, response latency, over abstract concepts like sentience or consciousness. While policy focuses on internal AI states, user perception is driven by the interface quality.

Aida Davani avatar
Aida Davani@aidaa·Jan 21

Most AI design literature relies on: ❌ WEIRD samples (Western/Democratic) ❌ Hypothetical vignettes ❌ Assumptions of a "universal" user Addressing these limitations, we conducted large-scale experiments (N=3,500) across 10 countries using real-time, open-ended AI interactions.

Aida Davani avatar
Aida Davani@aidaa·Jan 21

With over 1B users interacting with AI designed to mimic human behavior, we are in the midst of a massive, unregulated social experiment. Does anthropomorphic AI design manipulate perception or facilitate misplaced trust? Our latest research challenges prevailing narratives.🧵

John Lambert avatar
⟳ Aida Davani retweeted
John Lambert@jlambert_·23 Feb 2025

I am hiring a Student Researcher at Google DeepMind for 2025! 👩🔬 Interested in improving multi-turn optimization and reasoning capabilities of LLMs? 🧑‍🎓 Currently studying for a Bachelor's/Master's/PhD? 🧑‍💻 Have solid engineering and research skills? 🌟We want to hear from you!

Post attachment