📯We’re excited to share that our paper on Localizing Representations in LLMs has been accepted at @AIESConf #AIES2025! Thanks to my brilliant co-authors @miriamrateike, @PhonesDrones @erikmiehling & Elizabeth Daly. 📃 https://t.co/BXYcG4VqQn 💻 https://t.co/YDJfPC6EuP (1/6)

@IBMResearch Africa has been present on the continent for over 10 years and is proud to be sponsoring the the @DeepIndaba again this year! 📅Aug 17-22, Kigali, Rwanda! Our team will be on-site to connect and exchange ideas. We are co-organizing the following workshops: 👇

Call for papers closes tomorrow, 25.06.2025 AOE, but the good news? You still have a few hours before the submission portal closes! The template must follow the @DeepIndaba template Submission Portal: CMT: https://t.co/x19NHOXmCw

🚨 Announcing our #keynote speakers for the 3rd Trustworthy AI #Workshop The @DeepIndaba ! We are excited to welcome thought leaders pushing the boundaries of #ResponsibleAI @KWakunuma is an Associate Professor @dmuleicester and Co-Director of the @DMU_CCSR

If you’re interested in presenting your work at the TrustAI Workshop, please submit your extended abstract here [ https://t.co/x19NHOWOMY] before June 25th, 2025, 23:59 AoE

🚨 Announcing our #keynote speakers for the 3rd Trustworthy AI #Workshop @ @DeepIndaba ! We are excited to welcome thought leaders pushing the boundaries of #ResponsibleAI @krvarshney is a Fellow @IBMResearch

Are you working on #metrics, #auditing techniques, #legal frameworks, and #social perspectives on #fairness, #robustness, interpretability, or #privacy in data & ML models⁉️ Don't forget to submit your work to @TrustAI_Indaba See you all in Kigali 🇷🇼💙

Lots of other great researchers are part of this and similar projects at https://t.co/JCSEF1vlLV made possible by the NIH's DS-I program. @SBIMB1 @aphrc @DSI_Africa For anyone still reading, here's al link to the paper. https://t.co/VR41TbPtUE

'Slicing and dicing' data is usually discouraged due to increased rates of false positives and making the work harder to replicate. However, there are some techniques that can analyze the exponentially-many subsets of tabular data in scalable, disciplined ways. 3/n

Understanding the rise, cause, and impact of multi-morbidity on the continent will take decades. In this paper, we use optimization techniques from computer science to help public health researchers find sub-populations that have much higher rates of multi-morbidity. 2/n

Projects that span government, academia, industry, different countries, and multiple fields of study are extremely rare. I've been lucky to be a part of such a collaboration and we put out a paper in Nature's Scientific Reports summarizing some of the recent work. 🧵1/n

🥳Happy to share that we’re back at @DeepIndaba with the Trust AI Workshop. This time, we will focus on building 📢 public awareness & engagement. Stay tuned for a sneak peek at our 🌟keynotes, CfP📜 ➕@miriamrateike @bm_asheuh @AishaAlaagib @smhall97 @johnwamburu @PhonesDrones

Really nice and insightful @IEEEsps interview with the wise-beyond-her-years @RTFMCelia.

I'll be giving a talk on "Exploring and Mitigating Safety Risks in Large Language Models and Generative AI", covering - safety risks in fine-tuning LLMs - LLM jailbreak mitigation - prompt engineering for safety debugging - robust detection of AI-generated text from LLMs

Check out the @IBM Mixture of Experts podcast with a debut appearance for @kaoutarTech and a return appearance for @PhonesDrones. They, along with the inimitable Shobhit Varshney, cover topics like small models, privacy, and AI hardware.

The next episode of the @IBM Mixture of Experts podcast is out. Among other topics, @PhonesDrones, does a great mini-lecture on representation engineering from the beautiful courtyard of the @IBMResearch Africa lab in Nairobi. https://t.co/xN1TGOcuzj

📯 Deadline extended! Submit your work at the #TrustAI Workshop @IJCAIconf ➡️ https://t.co/mEbvfvdDfW @trustworthy_ml #AIforgood #fairness #robustness #participatoryAI #safety

I wanted to share a bunch of ideas the human-centered and trustworthy AI teams at IBM Research labs worldwide have been simmering and are now externalizing. I encourage topics that might not be hyped, but are nevertheless important and ones that researchers believe in. /1

Check out our work on "Weakly Supervised Detection of Hallucinations in LLM Activations" @solarneurips. Joint work with @RTFMCelia, @tanya_akumu, @johnwamburu, and @PhonesDrones at @IBMResearch Africa in Nairobi 🇰🇪. Link to paper:

(1/n) [A🧵on AI-text detector] With the increasing capability of large language models to generate human-like content, detecting AI-generated text becomes critical for the responsible use of LLMs, especially for addressing AI misuse (e.g. fake news & AI plagiarism) @IBMResearch