Zara Siddique @zarasiddique.bsky.social · 26/07/2025Loved giving my second tutorial on steering vectors at #CardiffNLPWorkshop Lots of enthusiastic participants! @cardiffnlp.bsky.social 042
Zara Siddique @zarasiddique.bsky.social · 14/07/2025#CardiffNLPWorkshop off to a flying start with talks from Jennifer Foster and Marianna Apidianaki @cardiffnlp.bsky.social 042
Zara Siddique @zarasiddique.bsky.social · 26/05/2025Pleased to say this has been accepted to ACL System Demos :) 0101
Zara Siddique @zarasiddique.bsky.social · 21/05/2025Come to my hackathon! Last one was super fun I promise 032
Zara Siddique @zarasiddique.bsky.social · 14/05/2025Shoutout to supervisors Liam Turner and Luis Espinosa-Anke and @cardiffnlp.bsky.social. I'm also interested in future collaborations on the topic so please message if you are interested :) 000
Zara Siddique @zarasiddique.bsky.social · 14/05/2025I highly encourage people to play around, you can get started in just a few lines. Here's a Colab notebook: tinyurl.com/yysmb45c Note that the results from this Colab won't be the best because it's using a smaller model to reduce loading times. I would recommend using at least a 7B.drive.google.comDialz Tutorial - Zara Siddique - KnitTogether 2025.ipynbColab notebook 100
Zara Siddique @zarasiddique.bsky.social · 14/05/2025As part of our validation, we see if we can reduce stereotypicality in outputs from Mistral 7B, using GPT-4o as a judge. There is a notable reduction compared to baselines and prompting, which is cool. 100
Zara Siddique @zarasiddique.bsky.social · 14/05/2025For those that are new to the topic, steering vectors are constructed using a set of paired sentences, where one elicits a 'positive' activation of neurons and the other elicits a 'negative' activation of neurons - by taking the difference, we isolate activations responsible for a certain 'concept'. 110
Zara Siddique @zarasiddique.bsky.social · 14/05/2025🚨 NEW PAPER ALERT 🚨 Dialz: A Python Toolkit for Steering Vectors ArXiv: arxiv.org/abs/2505.06262 Docs: cardiffnlp.github.io/dialz/ Repo: github.com/cardiffnlp/d... A Python package to help you create, apply and visualise steering vectors for anything you want - from sycophancy to bias. 2268
Zara Siddique @zarasiddique.bsky.social · 13/05/2025New friends! Old friends! Please register if you’d like 2 whole days packed with NLP fun 022
Zara Siddique @zarasiddique.bsky.social · 27/03/2025Love this take: "Society appears far more willing to critically examine and address bias in AI systems than confront human bias directly"theguardian.comCould AI help us build a more racially just society? | Sanmi KoyejoWe have an opportunity to build systems that don’t just replicate our current inequities. Will we take them? 010
Reposted by Zara SiddiqueDustin Wright @dustinbwright.com · 25/03/2025I am still in need of emergency reviewers for ARR this cycle for the computational social science track, please DM me if you have capacity 🙏 026
Zara Siddique @zarasiddique.bsky.social · 25/03/2025Do it! When interviewers ask me about them it’s usually a good sign that it’s a nice workplace. 110
Zara Siddique @zarasiddique.bsky.social · 13/03/2025The work presents the first systematic investigation of steering vectors for bias mitigation, and we demonstrate that SVE is a powerful and computationally efficient strategy for reducing bias in LLMs, with broader implications for enhancing AI safety. 010
Zara Siddique @zarasiddique.bsky.social · 13/03/2025Building on these promising results, we introduce Steering Vector Ensembles (SVE), a method that averages multiple individually optimized steering vectors, each targeting a specific bias axis such as age, race, or gender. 110
Zara Siddique @zarasiddique.bsky.social · 13/03/2025When optimized on the BBQ dataset, our individually tuned steering vectors achieve average improvements of 12.2%, 4.7%, and 3.2% over the baseline for Mistral, Llama, and Qwen, respectively. 110
Zara Siddique @zarasiddique.bsky.social · 13/03/2025We present a novel approach to bias mitigation in large language models (LLMs) by applying steering vectors to modify model activations in forward passes. We employ Bayesian optimization to systematically identify effective contrastive pair datasets across nine bias axes. 130
Zara Siddique @zarasiddique.bsky.social · 13/03/2025NEW PAPER 📜 Shifting Perspectives: Steering Vector Ensembles for Robust Bias Mitigation in LLMs ArXiv: arxiv.org/abs/2503.05371 GitHub: github.com/groovychoons... Extremely Unofficial Blog Post: zarasiddique.com/blog/shiftin...arxiv.orgShifting Perspectives: Steering Vector Ensembles for Robust Bias Mitigation in LLMsWe present a novel approach to bias mitigation in large language models (LLMs) by applying steering vectors to modify model activations in forward passes. We employ Bayesian optimization to systematic... 1175
Zara Siddique @zarasiddique.bsky.social · 10/03/2025Strongly encourage you to register for our free NLP workshop, previously had speakers from DeepMind, Microsoft, Amazon and top university NLP labs etc. and it’s looking like it’s going to be a great line up this year too. If you can’t make it, please share with others who may be interested! 011
Reposted by Zara SiddiqueCardiff NLP @cardiffnlp.bsky.social · 05/03/2025We've created a Cardiff NLP Starter Pack to make it easy to follow #NLP researchers at Cardiff Uni. 064
Reposted by Zara Siddique404 Media @404media.co · 29/01/2025OpenAI furious DeepSeek might have stolen all the data OpenAI stole from us 🔗 www.404media.co/openai-furio... 1073681929
Zara Siddique @zarasiddique.bsky.social · 26/01/2025Severance episode 2, Traitors final AND this It's a weekend of watching for me 🍿youtube.comJoy Buolamwini and Sam Altman: Unmasking the Future of AIYouTube video by Commonwealth Club World Affairs (CCWA) 010
Zara Siddique @zarasiddique.bsky.social · 26/01/2025Need to be spending less time on deepseek and more time on deep sleep 😴 020
Zara Siddique @zarasiddique.bsky.social · 23/01/2025Any thoughts to whether this would extend well to more traditional CompSci courses? 100
Reposted by Zara SiddiqueJoanne Boisson @joanneboisson.bsky.social · 20/01/2025Our paper on extraction of metaphoric analogies from literary texts will be presented in COLING ( Wed 11:00 , Atrium, poster) by @camachocollados.bsky.social and Luis Espinosa-Anke. Done with @zarasiddique.bsky.social , @hsuvas.bsky.social and @antypasd.bsky.social 164
Zara Siddique @zarasiddique.bsky.social · 21/01/2025If you ever find yourself getting worried about catastrophic AI risk, remember you can always just turn off the AI at the plug 000
Reposted by Zara SiddiqueThe Guardian @theguardian.com · 16/01/2025I knew one day I’d have to watch powerful men burn the world down – I just didn’t expect them to be such losers | Rebecca Shawtheguardian.comI knew one day I’d have to watch powerful men burn the world down – I just didn’t expect them to be such losers | Rebecca ShawElon Musk and Mark Zuckerberg’s desperation to be cool as they suck up to Donald Trump is so cringe it makes my skin crawl I don’t know if anyone else has noticed this but everything seems to be going down the tubes quite fast. And not fun tubes, like at… 992941891
Zara Siddique @zarasiddique.bsky.social · 16/01/2025arxiv.org/abs/2304.13734 This is an interesting one to addarxiv.orgThe Internal State of an LLM Knows When It's LyingWhile Large Language Models (LLMs) have shown exceptional performance in various tasks, one of their most prominent drawbacks is generating inaccurate or false information with a confident tone. In th... 110
Zara Siddique @zarasiddique.bsky.social · 31/12/2024Wrote a little blog post on my favourite EMNLP papers :) zarasiddique.com/blog/my-favo...zarasiddique.comZara Siddique 020
Zara Siddique @zarasiddique.bsky.social · 03/12/2024Had fun presenting my favourite #EMNLP2024 papers today at our secret reading group + bonus raccoon pics from Miami 🦝 Will follow up with favourite papers in blog post form soon! 020
Zara Siddique @zarasiddique.bsky.social · 20/11/2024Had an amazing time at #EMNLP2024 and excited to connect with other researchers on here :) 050
