Arslan Chaudhry
@arslan_mac
Research Scientist @GoogleDeepMind | Ex @UniofOxford, @AIatMeta, @GoogleAI
Our new work on studying the generalization of language models from in-context learning and finetuning is out. See the thread below by my amazing co-author @AndrewLampinen and read the paper here: arxiv.org/abs/2505.00661 Give us your feedback.
How do language models generalize from information they learn in-context vs. via finetuning? We show that in-context learning can generalize more flexibly, illustrating key differences in the inductive biases of these modes of learning — and ways to improve finetuning. Thread: 1/
Think you know Gemini? 🤔 Think again. Meet Gemini 2.5: our most intelligent model 💡 The first release is Pro Experimental, which is state-of-the-art across many benchmarks - meaning it can handle complex problems and give more accurate responses. Try it now →…
Still thinking about submitting to the continual foundational model development workshop at NeurIPS? We have extended the deadline to September 13 AoE. Please share it with others. sites.google.com/corp/view/cont…
Believe it, Nadeem. You just became Pakistan's first ever athletics Olympic champion. 🇵🇰🥹
I will be at the ICML in Vienna. If you are interested in continual learning in foundation models or other forgetting or adaptation related challenges in (V)LMs, come chat to me. I will be at/ around the @GoogleDeepMind booth.
We're looking for an exceptional junior researcher in AI/ML with strong interests in diversity, equity and inclusion to fill this exceptional role funded by a generous donation from @GoogleDeepMind. Deadline for application 5 July 2024.
🔎New Senior Postdoc Research Associate Position! We’re looking for someone to be a role model, ambassador & point of contact for emerging talent in the field of #AI or #MachineLearning. 🔗 For more details & to apply, visit: my.corehr.com/pls/uoxrecruit… #postdoctoralresearch
Very well done @usmananwar391 and coauthors.
I’m super excited to release our 100+ page collaborative agenda - led by @usmananwar391 - on “Foundational Challenges In Assuring Alignment and Safety of LLMs” alongside 35+ co-authors from NLP, ML, and AI Safety communities! Some highlights below...
I have no doubt about the positive intentions behind this, but my fear is that it's yet another policy that will benefit kids with already exceptional access to education and mentorship. How about a specific track for work on AI that aims to equalize educational opportunity?
Gemini has achieved a classification accuracy of 90% on MMLU (with a combo of greedy sampling and a majority vote of 32 COT samples). This is remarkable! For reference, expert-level perf, defined at 95th percentile human test-taker accuracy in *each subject*, would be ~89.8%.
Exciting times, welcome Gemini (and MMLU>90)! State-of-the-art on 30 out of 32 benchmarks across text, coding, audio, images, and video, with a single model 🤯 Co-leading Gemini has been my most exciting endeavor, fueled by a very ambitious goal. And that is just the beginning!…
Our team in @GoogleDeepMind Mountain View is looking for a Research Scientist/ Research Engineer. If you are interested in working on multimodal foundation models and their adaptability over time, come chat with me. I will be attending ICML and my schedule is quite flexible.
Another AI paradox: people are excited about LLMs, some even think that AGI is just around the corner. But some students are depressed how they can still get a PhD. Is it becoming pointless? Some personal notes on this. (1/8)
Our work on a new large-scale benchmark composed of 30 years of computer vision research to explore knowledge accrual in continual learning systems.
Introducing NEVIS’22, a new benchmark developed using 30 years of computer vision research. This provides an opportunity to explore how AI models can continually build on their knowledge to learn future tasks more efficiently. ➡️ dpmd.ai/3VqnkHc