AI at Meta’s Post

The next logical step after getting good with single word prediction. How far will we go down the branch prediction path?

Something very interesting in the field of NLP!!

This multi-token prediction approach sounds revolutionary for LLMs! Predicting multiple words at once could significantly enhance both efficiency and capability. But how does this method impact the accuracy of predictions in more complex contexts? Releasing pre-trained models for code completion is a fantastic step for collaborative progress. However, as we advance, how do we ensure these models are responsibly used and continuously improved? Exciting times ahead for the research community and AI development! #MultiTokenPrediction #AIResearch #FutureOfAI

Worth the read: Gao, Yunfan, et al. Retrieval-Augmented Generation for Large Language Models: A Survey. 1, arXiv:2312.10997, 18 Dec. 2023. arXiv.org, https://lnkd.in/e5jVfUYC.

🧠📙 𝐀 𝐦𝐚𝐭𝐡𝐞𝐦𝐚𝐭𝐢𝐜𝐚𝐥 𝐩𝐞𝐫𝐬𝐩𝐞𝐜𝐭𝐢𝐯𝐞 𝐨𝐧 𝐓𝐫𝐚𝐧𝐬𝐟𝐨𝐫𝐦𝐞𝐫𝐬 🚀💥 - an interesting paper about math foundations of #transformers by BORJAN GESHKOVSKI, CYRIL LETROUIT, YURY POLYANSKIY, AND PHILIPPE RIGOLLET 🎯 Transformers play a central role in the inner workings of large language models. 🎯 We develop a mathematical framework for analyzing Transformers based on their interpretation as interacting particle systems, which reveals that clusters emerge in long time. 🎯 Our study explores the underlying theory and offers new perspectives for mathematicians as well as computer scientists. #llms #largelanguagemodels #deeplearning

🤯LLEMA is out 🤯 LLEMMA a new open source LLM specifically designed to solve mathematical problems, surpass Minerva by Google and other competitors on an “equi-parameter basis.”🔥 This means that LLEMMA-7B outperforms Minerva-8B, and LLEMMA-34B is nearly on par with Minerva-62B on a 4-shot Math Performance. 👏 Link to the research paper in the comments and here https://lnkd.in/gpbiEZUz #llema #llm #mathematics #opensource #datascience #news #minvera #maths #generativeAI #genAI #machinelearning #news

Microsoft'tan kurumsal üretkenlik odaklı yeni yapay zeka sistemi: SpreadsheetLLM Encoding Spreadsheets for Large Language Models (arxiv.org)

draft^media002_Transculpting As the research transitioned from theoretical to practical, the progress started anchoring itself around alternative terminologies. Transculpting is a product of this speculative process. It is the act of defining neural style transfers as tangible physicalities. Therefore the transfers of style surpass the bidimensional threshold and manifest in 3D space. This episode of the thesis worked towards developing and correlating a proper partnership of exchange between architectural machine learning and analog meaning-making. Académie Libanaise des Beaux-Arts Guided by: SAMER EID

The title of our latest pre-print is: "Evaluating the Effectiveness of GPT-4 Turbo in Creating Defeaters for Assurance Cases". The authors of that preprint are: Kimya Khakzad Shahandashti, Mithila Sivakumar, Mohammad Mahdi Mohajer, Alvine Boaye Belle, Song Wang, and Timothy Lethbridge. Our pre-print is available on ArXiv. You can access it by using the following link: https://lnkd.in/dsvuEBKj The abstract of our paper is the following: "Assurance cases (ACs) are structured arguments that allow verifying the correct implementation of the created systems’ non-functional requirements (e.g., safety, security). This allows for preventing system failure. The latter may result in catastrophic outcomes (e.g., loss of lives). ACs support the certification of systems in compliance with industrial standards e.g. DO-178C and ISO 26262. Identifying defeaters —arguments that challenge these ACs — is crucial for enhancing ACs’ robustness and confidence. To automatically support that task, we propose a novel approach that explores the potential of GPT-4 Turbo, an advanced Large Language Model (LLM) developed by OpenAI, in identifying defeaters within ACs formalized using the Eliminative Argumentation (EA) notation. Our preliminary evaluation assesses the model’s ability to comprehend and generate arguments in this context and the results show that GPT-4 turbo is very proficient in EA notation and can generate different types of defeaters.".