Eran Dahan’s Post

משרה מאתגרת, מעניינת והזדמנות לעבוד עם אחד האנשים המעניינים שאני מכיר!

I was 𝐭𝐡𝐫𝐢𝐥𝐥𝐞𝐝 and 𝐡𝐨𝐧𝐨𝐫𝐞𝐝 to speak at 𝐂𝐲𝐛𝐞𝐫 𝐚𝐧𝐝 𝐀𝐈 𝐖𝐞𝐞𝐤 𝟐𝟎𝟐𝟒 at Tel Aviv University. It was an incredible experience to share insights on the fascinating intersection of 𝐜𝐲𝐛𝐞𝐫𝐬𝐞𝐜𝐮𝐫𝐢𝐭𝐲, 𝐚𝐫𝐭𝐢𝐟𝐢𝐜𝐢𝐚𝐥 𝐢𝐧𝐭𝐞𝐥𝐥𝐢𝐠𝐞𝐧𝐜𝐞, 𝐚𝐧𝐝 𝐝𝐢𝐠𝐢𝐭𝐚𝐥 𝐢𝐧𝐧𝐨𝐯𝐚𝐭𝐢𝐨𝐧. I loved to meet with new and old friends with joint areas of interest and to connect with industry leaders, decision-makers, and experts from around the globe, it was truly inspiring. The discussions and ideas exchanged have only reinforced the importance of these fields in shaping our future. A big thank you to everyone who attended and engaged in the conversation.Looking forward to more opportunities to collaborate and innovate together! #AIWeekTLV #AIWeek24

English below צעד נוסף בפיתוח מודלי שפה גדולים (llm) בעברית, מפא"ת DDR&D (MAFAT) - Directorate of Defense Research & Development , בשיתוף חברת Webiks, מנגישה לציבור מאגר מתויג (dataset) של כ-300,000 משפטים ב-NLI . המאגר מבוסס על multinli , שעבר תרגום אוטומטי ע"י מודל שפה גדול - Gemini ובחינה לוגית ולשונית מעמיקה. בעבודה ווידאנו שהתרגום לא משפיע/משנה את התיוג המקורי, ושהמשפט המתורגם נשמר בשפה טבעית. כמו כן בחנו מהי הדרך האופטימלית כיום לתרגם משפטים. היות ש-multinli מורכב ממספר מקורות מידע (ג'אנרים) שונים, בוצעו תרגום ובדיקות אימות נפרדות לכל ז'אנר, ונופו ז'אנרים שתרגומם לא היה טוב מספיק. המאגר המתורגם, התרגומים לכלל מאגר הבסיס - multinli, וה-prompts שבהם השתמשנו לתרגום, כמו גם תיעוד התהליך מפורט זה זמינים ב-github שלנו: https://lnkd.in/erEEKccT וב-huggingface : https://lnkd.in/eG7Zx26i. לעוד פרטים, מפנה לפוסט של Tal Geva, https://lnkd.in/evE59Pce A further step in the development of large language models (LLMs) in Hebrew, DDR&D (MAFAT) - Directorate of Defense Research & Development (MAFAT), in collaboration with Webiks , has made available to the public a dataset of about 300,000 sentences in NLI. This dataset is based on multinli, which has undergone automatic translation by a large language model - Gemini, and has been thoroughly checked for logical and linguistic accuracy. In this work, we ensured that the translation does not affect/change the original tagging and that the translated sentence remains in natural language. We also examined the most optimal way to translate sentences currently.Since multinli consists of various sources of information (genres), translations and verification tests were performed separately for each genre, and genres whose translations were not good enough were omitted. The translated dataset, the translations for the entire base dataset - multinli, and the prompts used for translation, as well as detailed documentation of this process, are available on our GitHub: https://lnkd.in/erEEKccT And on Hugging Face: https://lnkd.in/eG7Zx26i For more details, refer to the post by Tal Geva https://lnkd.in/evE59Pce

*Digital Transformation: A Deep Dive into the Future* As the holiday season concludes, my reflections on digital transformation have intensified. Let's journey back a decade. The ImageNet challenge was our benchmark. Back then, the notion of algorithms matching human prowess in image classification seemed ambitious. Yet, we surpassed that threshold. Fast forward to the present(Gartner image below), and the realm of natural language processing stands transformed. By 2023, large language models (LLMs) have not only grappled with the intricacies of human language – with its sparse data and unstructured nuances – but have started defining myriad applications. Drawing from historical precedents, if I were to prophesize: 1️⃣ LLMs will permeate uncharted territories. We'll witness their application in specialized fields such as RF communication protocols, satellite data processing, and complex autonomous systems, transitioning from proof-of-concept to industry standards. 2️⃣ The rise of multimodal models is imminent. These models will bridge digital and physical worlds. Envision an LLM seamlessly integrated into augmented reality glasses or IoT devices, continuously interfacing with our environment. This synthesis could redefine human-machine interaction. There's an age-old maxim, "Prophets were once deemed foolish." Venturing a prediction, it is evident that we're on the precipice of transformative change. However, with great power comes great responsibility. Enterprises and governmental bodies must evolve. It's paramount to establish dedicated centers of excellence, akin to the previously instituted Cloud Center of Excellence (CCOE), now focusing on *Machine Learning - an MLCOE*. The challenges are multifaceted: cybersecurity, financial oversight, and operational integration. Years back this is what I did in the Directorate of Defense Research & Development and looking back I think we got very far, it was a fascinating, annoying journey with a lot of self learning and challenges. Digital transformation isn't a future concept; it's our present reality. The next decade promises exponential growth, and preparedness is our only recourse. *If you are part of such a big organization. Act now and stay relevant.* Certainly. Here are some professional and relevant hashtags for your post: #DigitalTransformation,#MultimodalAI,#MLCOE,#MilitaryInnovation,#DefenseAI,#StrategicTech

🌟Exciting Opportunities with LLM in the Defence Industry🌟 In developing the next wave of R&D for defence, especially within the Directorate of Defense Research & Development the power of Large Language Models (LLM) cannot be overstated. They stand out in processing vast data, improving situational awareness, and strengthening autonomous systems. However, we must confront substantial challenges. Issues like privacy, controlled insight sharing across different domains, on-the-fly context acquisition as in RAG, and tailored data fine-tuning are critical. Addressing these is vital to unleash the full capabilities of LLM within the defence sector. Are you, like me, intrigued by this advanced technology? Eager to surmount these obstacles and lead groundbreaking progress in this field? Let’s embark on this exhilarating journey! If you're filled with innovative ideas and solutions, your insights are welcome. PM me, and let’s discuss how we can join forces to make a meaningful impact with LLM in defence. #LLM #DefenceTechnology #Innovation #Collaboration

👉 Navigating through the challenges of training large language models and vast architectures in machine learning is *hard*. 👉 From computational requirements to model interpretability, the challenges are many, but so are the rewards. High-Performance Computing (HPC) applications further expand this, demanding precision, scalability, and a profound understanding of data dynamics. 👉👉 On October 15th, as part of the AI Summit at NVIDIA in Tel Aviv, NVIDIA professionals will host a workshop 'Model Parallelism: Building and Deploying Large Neural Networks'. The potential of large model architectures and their deployment nuances will take center stage. 👉 I will definitely be present, to both share insights and learn. Let's explore the mutual potential collaborations in this captivating domain. Hope to see many familiar faces and meet there! #AISummit #NVIDIA #MachineLearning #HPC #ModelTraining

Last year was something very different for me, marking a turning point in my professional life. After 3 fulfilling years as the Head of the AI branch at IMOD, where I had the privilege to lead an inspiring team of officers, professionals, and team players. Together, we shaped our IMOD AI strategy, developed new techniques, algorithms, and capabilities that pushed the boundaries of what's possible. I then had the opportunity to shift my focus to personal growth, dedicating myself to my Ph.D. in Electrical Engineering, specializing in AI and computer vision. This time was also full by a well-earned break with my family, filling our lives with new experiences, love, laughter, and precious memories. (evidence below :) ) As the new chapter begin in a few days, I'll be starting as the Head of Digital and Data Branch. The road ahead is filled with opportunities to explore emerging and transformative technologies like Generative AI, cloud security, quantum computing, and autonomy. Our goal is to invest in our core architecture, ensuring we're fast learners and can securely adopt these innovations. This will not only keep us relevant but also empower us to lead in the relentless race of defense. Here's to the next adventure! #NewBeginnings #Innovation #Leadership #Technology #AI #DigitalTransformation

ממליץ מאוד לכל מי שבטחון-בינה מלאכותית הוא תחום העניין שלו!