We introduce h4rm3l, a language, a synthesizer, and a scalable and explainable dynamic LLM red-teaming toolkit. h4rm3l found > 2.6k new jailbreak attacks targeting @OpenAI, @AIatMeta, and @AnthropicAI LLMs. 📝 https://t.co/BjgHgW66Ru 🌐 https://t.co/GcKvDQcV4g 🧵1/6 👇🏾

Announcing DataGemma, a set of open models that utilize Data Commons through Retrieval Interleaved Generation (RIG) & Retrieval Augmented Generation (RAG) to ground LLMs in real-world data for fact-checking, responsible AI development & more. Read more at https://t.co/0dGE6Qqf9c

Mitigating Hallucination in LLMs This paper summarizes 32 techniques to mitigate hallucination in LLMs. Introduces a taxonomy categorizing methods like RAG, Knowledge Retrieval, CoVe, and more. Provides tips on how to apply these methods and highlights the challenges and

Excited to participate in the @PistoiaAlliance Digital Transformation in R&D conference today! I’ll be joining a panel to discuss federated technologies and what @lifebitAI is doing to enable federated data access, analysis and ML across multi-omics, EHR and imaging data.

In a major scientific breakthrough, the latest version of #AlphaFold has been recognised as a solution to one of biology's grand challenges - the “protein folding problem”. It was validated today at #CASP14, the biennial Critical Assessment of protein Structure Prediction (1/3)

🎉 Papers with Code partners with arXiv! Code links are now shown on arXiv articles, and authors can submit code through arXiv. Read more: https://t.co/kO6zhWAWGH

@markus_zechner @omarsar0 I recommend https://t.co/PZ5FOscj67 on NLP in general. Linguistics wise, I am reading Emily M. Bender's two books morphology and syntax, and semantics and pragmatics. They are not easy to absorb but are very interesting. I'd love to see more linguistics recommendations as well.

Fascinating insight: Trained solely on unsupervised language modeling, the Transformer's attention mechanism recovers high-level structural (folding) and functional properties of proteins! Blog: https://t.co/ClWXxxNbpW Paper: https://t.co/0oK7coEMV6 Code: https://t.co/3z3CVQ51kS

Neural Architecture Search without Training

Excited to announce the Hateful Memes Challenge, a new dataset and competition for vision and language, focusing on multimodal understanding and reasoning: https://t.co/pbkqLQZOyv "Mean" memes shown here as an illustration of the problem. Some highlights: (1/7)

#AI & #DrugDiscovery - a great overview of the @JLABS & @Innovation_F Cambridge event featuring our #MachineLearning engineer @0_rishabh 🙌🚀

We are absolutely thrilled to collaborate with @sanogenetics to further our joint mission of advancing knowledge, making #genomics #research faster & more patient-centric, while helping to democratise access to the latest #bioinformatics techniques.

Classifiers are secretly energy-based models! Every softmax giving p(c|x) has an unused degree of freedom, which we use to compute the input density p(x). This makes classifiers into generative models without changing the architecture. https://t.co/ObcbI8kZdz

Senior #ML Engineer at @Lifebit, @0_rishabh, presenting at 'Science First: Personalising #Healthcare Through #AI', organised by @JLABS & @IF_Cambridge. Rishabh elaborates on how to make data AI-ready by improving the #quality, #accessibility & #Interoperability of #data🚀

“Now What?” John J. Hopfield Hopfield is a theoretical physicist, molecular biologist, neuroscientist, and AI researcher. His broad background led him to publish a 5-page (unrefereed) paper on what is now known as Hopfield Networks, his most cited work. https://t.co/2XppOx2r8M

Trends 📈 - track the popularity of deep learning frameworks for paper implementations. Current share in Q3 2019: PyTorch 38% (up 6%), TensorFlow 22% (down 2%), other 39% (down 3%) https://t.co/NgItkEMPEx

The Hamiltonian formalism is fundamental to physics and possesses many properties useful for machine learning. Our new papers show how to bring these properties to neural networks:

Really excited to be sharing the results of our + others' research into language models and their social impacts, as well as what we've learned on publication norms in the past 6 months. https://t.co/vRcFryFdUg Direct link to the report:

Applications are now open for our 2020 Research Scientist and Engineering internships! We offer hands-on experience working collaboratively on projects that push the frontiers of AI and science across a wide variety of teams. Find out more or apply:

Introducing Deepmind: The Podcast, a project we’ve been working on for the last twelve months.