Very excited to present Minerva🦉: a language model capable of solving mathematical questions using step-by-step natural language reasoning. Combining scale, data and others dramatically improves performance on the STEM benchmarks MATH and MMLU-STEM. goo.gle/3yGpTN7

Read a bit about Grokking recently. Here's some learnings: "Grokking" is a curious neural net behavior observed ~1 year ago (arxiv.org/abs/2201.02177). Continue optimizing a model long after perfect training accuracy and it suddenly generalizes. Figure:

Are large pre-trained models nothing more than stochastic parrots? Is scaling them all we need to bridge the gap between humans and machines? In this new opinion piece for Nautilus Magazine, I argue that the answer lies somewhere in between. 1/14 nautil.us/moving-beyond-…

How can we rigorously evaluate the anticipated risks of language models (LMs)? Our team identifies six “characteristics” of LM-generated text which can help guide the design of more rigorous benchmarks. Learn more: dpmd.ai/dm-lm-benchmark 1/

Google’s LaMDA has a welterweight 137B parameters. It trained longer than GPT-3 though: 1024 TPUv3 cards for 57.7 days. This would cost around $6M in Amazon dollars with an equivalent number of A100s. Better start saving up those free AWS credits, grad students.

Facebook Instagram Wikipedia Discover how our No Language Left Behind project is driving inclusion through the power of AI translation: bit.ly/3ut2Oe6

Introducing our work Google AI CALM: Confident Adaptive Language Modeling 🧘 Large Language Models don't need their full size for every generated token. We develop an Early Exit framework to significantly #accelerate decoding from #Transformers! 🔗: arxiv.org/abs/2207.07061 🧵1/

How AI allows us to paint pictures with words. New AI applications can generate images based on natural language prompts provided by human users. By @tara_wilfred TU Dublin / OT Baile Átha Cliath rte.ie/brainstorm/202…

Developers: it's a simple feature, users will understand it. Users:

Can AI help to increase access to all languages? The No Langage Left Behind AI project is looking to create an effective and efficient way to translate between 200 languages. By @tara_wilfred TU Dublin / OT Baile Átha Cliath rte.ie/brainstorm/202… #WorldTranslationDay

Is it just me or are we all curious to know Tesla’s supercomputer vs. Riken’s Fugaku specs? Although, what intrigues me more is that Elon Musk’s ventured into #QuantumComputing. Now that’s something to look forward to. #tesla #ai #elonmusk #NeuralNetworks

(1/4) Today, we’re sharing how we built an AI system that can interpret Show More and Show Less signals from select posts to predict the types of content people want to see in their Feeds. Learn more about how it works: bit.ly/3SDM123

#AICS2022 #AICS22

Clips from my Research Presentation on Author Gender Identification considering Gender Bias at #AICS2022

Celebrating the End of the Autumn Trimester.

Thank you so much! Grateful, honoured, and humbled! Here's to more innovation, collaboration, and breaking barriers in tech! 👩🏻‍💻 #tech #mllabs #tud #tudublin TU Dublin / OT Baile Átha Cliath ML-Labs #dita #dita2024 #MachineLearning #ArtificialIntelligence #research

I was invited as a Panel Speaker for the “AI for Impact: Shaping Ethical AI Futures” event organized by Women in AI & EY. Turned out to be engaging, discussing the direction of AI, its ethical implications, bias, the EU AI Act, GenAI & Creators, etc. instagram.com/reel/DCDQ1MnCo…