←
NeuraSearch Laboratory237 posts
NeuraSearch Laboratory avatar

NeuraSearch Laboratory

@NeuraSearchLab

Neuroscience. Brain Computer Interaction. Search. Information Retrieval. NLP. AI. Deep Learning. Data Science. Data Analytics. University of Strathclyde.

◷ Joined November 2018▧ 57 Photos and videos
NeuraSearch Laboratory avatar
NeuraSearch Laboratory@NeuraSearchLab·Sep 17

📢AI, Information Overload, and What Comes Next Join us- 3 Dec 2026 at the University of Strathclyde, for a research workshop on AI, IR & info overload. Hear findings from the OMINO project and discuss emerging challenges and what comes next. 🎟 Register: eventbrite.com/e/ai-informat…

NeuraSearch Laboratory avatar
NeuraSearch Laboratory@NeuraSearchLab·Sep 7

📢 “Growing up with AI” - free public event on how kids/young people are using ChatGPT & what it means for learning, memory & independent thinking. 🗓️ 27 Oct, 3–5pm 📍 Strathclyde (hybrid) 👥 Parents, guardians, educators + anyone interested (16+) Reg: eventbrite.co.uk/e/199902416…

NeuraSearch Laboratory avatar
NeuraSearch Laboratory@NeuraSearchLab·18 Nov 2025

Dr Elizabeth Bent shared exciting insights on responsible AI development. Great news: UK government plans to expand AI budget investment over the next 4 years! 📷

NeuraSearch Laboratory avatar
NeuraSearch Laboratory@NeuraSearchLab·5 Nov 2025

📑 Read the paper: arxiv.org/abs/2509.08853 🖥️ Code & Data: github.com/CIS-PHAWM/POW #NeuraSearchLab #EMNLP2025 #ResponsibleAI #LLMs #AIBias #NLProc #AIEthics #PHAWM #RAI #StrathclydeUni #DemocraticAI

NeuraSearch Laboratory avatar
NeuraSearch Laboratory@NeuraSearchLab·5 Nov 2025

DeepSeek models showed the most restrictive boundaries at around 1% coverage, while Gemini 2.0 Flash was the most expansive at 67%. Crucially, most models struggle to articulate authoritarian or right-libertarian perspectives.

NeuraSearch Laboratory avatar
NeuraSearch Laboratory@NeuraSearchLab·5 Nov 2025

generating over 17,000 essays across 62 political propositions. What they found was striking: while most models default to economically left and socially liberal positions, their willingness to express diverse viewpoints varies dramatically.

NeuraSearch Laboratory avatar
NeuraSearch Laboratory@NeuraSearchLab·5 Nov 2025

Within this work, they explore political bias in LLMs by moving beyond traditional point estimates to map the actual boundaries of what different models are willing or unwilling to say. Using their PRISM methodology, they audited 28 LLMs from eight providers.