Amina Abdullahi
563 posts
@amilah_dul
CS PhD student @BrownCSDept | Biomedical AI | IR | NLP.
🚨 Reasoning models can “self-jailbreak”: they recognize a request is harmful, invent a reason why it’s fine, then help with it. We found that after training on benign math/code reasoning, models emergently start to reason themselves out of safety alignment. 🧵👇
Congratulations to David Adelani (@davlanade), Core Academic Member at Mila and Canada CIFAR AI Chair, who has been selected for the prestigious AI2050 Early Career Fellowship program by @schmidtsciences. This recognition highlights the crucial importance of his work in developing more equitable and inclusive AI models. mila.quebec/en/news/profes…
Together AI 🤝@CollinearAI Introducing TraitMix, Collinear’s simulation product empowering teams to generate persona-driven AI agent interactions. 🔌Plug these interactions into your workflows and evaluate their effectiveness with Together Evals. Details: bit.ly/43GHJhR
🤔Ever wonder why LLMs give inconsistent answers in different languages? In our paper, we identify two failure points in the multilingual factual recall process and propose fixes that guide LLMs to the "right path." This can boost performance by 35% in the weakest language! 📈
🧵 Multilingual safety training/eval is now standard practice, but a critical question remains: Is multilingual safety actually solved? Our new survey with @Cohere_Labs answers this and dives deep into: - Language gap in safety research - Future priority areas Thread 👇
Best Social Impact Paper #ACL2025NLP
Happy to share our paper was accepted at CVPRw Computer Vision the Wild 🎉
