🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

Excited to be discussing #research I'm doing on #LLMs and #misinformation Oxford Internet Institute and meedan in a free webinar Friday. Also talking about #funding opportunities from ESRC Digital Good Network . Please join us

My big ask of 2024: how can we design systems of human feedback which are sustainable, truly participatory (not extractive), privacy-preserving & inclusive of pluralistic values? Lots of attention on open models but what about post-training data? A passionate🧵+ call to action👇

New article! How can we better align Large Language Models with diverse humans? ask Hannah Rose Kirk and Scott Hale (Scott Hale) Oxford Internet Institute OxfordSocialSciences @UnioOxford oii.ox.ac.uk/news-events/ho… 1/3

Published in Nature Machine Intelligence today, our new article explores the trade-offs of personalised alignment in large language models ⚖️ Personalisation has potential to democratise decisions over how LLMs behave, but brings its own set of risks... nature.com/articles/s4225…

#AI is often fine-tuned to the preferences of a small group; #personalization seems an obvious way to be more #inclusive. It probably is, to a point. In this Nature Machine Intelligence paper we examine both the benefits and #risks of personalizing #LLMs. nature.com/articles/s4225…

Today we're launching PRISM, a new resource to diversify the voices contributing to alignment. We asked 1500 people around the world for their stated preferences over LLM behaviours, then we observed their contextual preferences in 8000 convos with 21 LLMs arxiv.org/abs/2404.16019

🚨Our latest The Web Conference paper!🚨 We study global #news coverage based on 60 million news published in 2020 across 124 countries, identifying 4,357 key news events, with Przemyslaw Grabowicz Scott Hale Mattia Samory David Jurgens is now on BlueSky only Ethan Zuckerman. Paper: arxiv.org/abs/2405.00280 🧵1/5

I'm excited to share that our Demo paper, "SynDy: Synthetic Dynamic Dataset Generation Framework for Misinformation Tasks," has been accepted to #SIGIR #SIGIR2024! With michael shliselberg Ashkan Kazemi Scott Hale: Preprint is now available. 🧵 arxiv.org/abs/2405.10700

🚨ANNOUNCEMENT🚨 Meedan is launching the 2024 #InvestigativeJournalismFellowship today! We’re working with Agência Pública, مصدر عام | The Public Source, and NIRIJ to promote #ElectionIntegrity through investigative reporting. Read about it: ow.ly/nnj750S7Gwi #JournalismForDemocracy

🌎Introducing LINGOLY, our new reasoning benchmark that stumps even top LLMs (best models only reach ~35% accuracy)🥴 In a colab between University of Oxford, Stanford University and UK Linguistic Olympiad puzzle authors, we stress test LLMs on over 90 low-resource and extinct languages...

‼️New preprint: Scaling laws for political persuasion with LLMs‼️ In a large pre-registered experiment (n=25,982), we find evidence that scaling the size of language models yields sharply diminishing persuasive returns: arxiv.org/abs/2406.14508 1/n

Two great fully funded #PhD opportunities. #studentship . One on #elections and #misinformation and one on technology-facilitated gender-based violence and #hate

My twitter feed this weekend is full of people praising Roberta and recommending this internship opportunity 😂 I agree so I’m adding one more tweet to this list!

Check out my blog post on meedan's recent work on using LLMs for running a multi language classification service called ClassyCat🐈‍⬛ details below🧵

Great advice

Amazing work Hannah Rose Kirk . It's been amazing working together on this, and to see you grow as leader and scholar. Congratulations 🎉 cc:Oxford Internet Institute

Humans are social creatures and AI agents are increasingly capable of relationship-building behaviors. What does this mean for safe and aligned AI? 🧐 We ask and answer this question in our new preprint w/ Iason Gabriel summerfieldlab @summerfieldlab.bsky.social, Bertie Vidgen, Scott Hale FAQs in 🧵

Are LLMs biased when they write about political issues? We just released IssueBench – the largest, most realistic benchmark of its kind – to answer this question more robustly than ever before. Long 🧵with spicy results 👇

📈Out today in PNASNews!📈 In a large pre-registered experiment (n=25,982), we find evidence that scaling the size of LLMs yields sharply diminishing persuasive returns for static political messages.  🧵:

Why do human–AI relationships need socioaffective alignment? As AI evolves from tools to companions, we must seek systems that enhance rather than exploit our nature as social & emotional beings. Published today in nature Humanities & Social Sciences! nature.com/articles/s4159…