David Talby’s Post

Introducing the Yarn-Llama-2-13b-128k model, a groundbreaking creation by the brilliant minds at NousResearch and EleutherAI! 🌟🔍 This model, with its impressive 128k context length, opens up endless possibilities for Document AI. Where it can be used for understanding and processing longer and more complex documents effortlessly! 🚀 Its commercial license, making it accessible for various industries and applications, alongside with the capability to handle extensive context, are a game-changer for tasks that require deep comprehension and analysis of documents, from legal documents to scientific research papers. 📚 The magic behind this innovation lies in the cutting-edge 𝗬𝗮𝗥𝗡 scaling techniques, pushing the boundaries of context length while maintaining top-notch performance. The models have 𝘀𝗶𝗺𝗶𝗹𝗮𝗿 𝗽𝗲𝗿𝗳𝗼𝗿𝗺𝗮𝗻𝗰𝗲 to the base LLaMA 2 models on the Open LLM benchmarks while scaling 𝗰𝗼𝗻𝘁𝗲𝘅𝘁 𝗹𝗲𝗻𝗴𝘁𝗵 𝗱𝗶𝗿𝗲𝗰𝘁𝗹𝘆 𝘁𝗼 𝟭𝟮𝟴𝗸. The implementation for YaRN rotary embeddings can be found here: https://lnkd.in/eqc4XXMS #llama2 #128k #contextlength #documentautomation #documentAI #gpt #huggingface

while it took 45 minutes to complete #LLM fine-tuning, the prep work on dataset took a entire day! more details comming here , #inspirellm is here to stay ! #genai #ai slow steps towards atleast general AI not #AGI https://lnkd.in/gaSjVifd

Reinforcement learning is undoubtedly one of the most incredible implementations of AI for the future. Here are two implementations for the same scenario: DQN and PPO. https://lnkd.in/evuPzkZv https://lnkd.in/eyBc6nXA And here a more exhaustive explanation of DQN https://lnkd.in/et94gafD

👀 👇 Don't miss this great conversation between Duke University School of Medicine Dean Mary E. Klotman M.D. and Duke AI Health Director Michael Pencina, PhD, about the "two revolutions" in data and technology that are expanding the promise of #AI for healthcare: https://lnkd.in/esyeGgRy #DukeAI

Exciting news in the world of AI! 🚀 A new method called YaRN is shaking up the game for large language models (LLMs). It supercharges context capabilities by using the RoPE technique, allowing for extended context up to 128k tokens. 🧠 Potentially a game-changer for accommodating lengthy texts and message histories and I am particularly interested to see how this will affect the attention mechanisms, while keeping the original model structures intact. 🌐 #AI #Innovation #YaRN #LanguageModels #Transformers #LLM #GenAI https://lnkd.in/ge5_p22z

The DMRC took centre stage at the Evergreens stage in a captivating session on AI and Society during this year’s 2023 Woodford Folk Festival. Moderated by DMRC Director Professor Daniel Angus, panellists included Distinguished Professor Jean Burgess, Associate Director of the national ARC Centre of Excellence for Automated Decision-Making and Society, Dr Aleesha Rodriguez, PhD, and PhD student Daniel Whelan-Shamy. This panel ignited thought-provoking dialogue on AI's impact across multiple sectors, and our daily lives. The discussion explored generative AI, algorithmic advertising, banking, government systems, retail, and education, unravelling the profound impact of automation. Distinguished Professor Burgess underscored the importance, stating, "The impact of automation in our daily lives is immense. It's a discussion that's increasingly important." This marks DMRC's second year participating in Woodford Folk Festival discussions, showcasing our commitment to addressing pertinent issues at the intersection of technology and society. Read more here 👇 #QUTResearch

⏰ Only one day to go until the Dimensions webinar ⏰ ‘What are Trust Markers and how does Dimensions Research Integrity use them to strengthen research integrity?’ Take a deep dive into Trust Markers and discover how Dimensions Research Integrity uses #AI to help you 📊 Compare performance against your organization’s peers and against global trends 📏 Measure the effects of targeted interventions 🛡️ Protect your reputation as a producer, publisher or funder of ethically compliant research 🔍 Identify the areas of research practice that need the most attention ✍️ Register here: https://lnkd.in/ejqtncmj #ResearchIntegrity #MindtheTrustGap #Dimensions

2023 Round-up: We asked our researchers what their favourite articles were this year: What stuck with them, and what made them think? Our answers are in the link below - but we'd love to know what articles about AI everyone else has enjoyed this year? https://lnkd.in/ebf2XnNP

It is a bizarre news on AI-generated materials in a peer reviewed scientific paper. The paper and the news, both are linked in the video description.  https://lnkd.in/g4nVxi8m #generativeai #generativeart

Are Large Language Models such as Chat-GPT or AI Art Systems, such as Midjourney, topics that you want to know more about? 🎨⌨️ On the 28th of August, leading scientists and industry shapers are coming to EPFL to discuss Generative AI and share their thoughts and perspective on this latest game-changing tech, it's effect on healthcare, the job market, start-ups, new biopharmarceuticals, and the democratization of it all 🧠 More information ⬇️