AccGPT is an on-going project to integrate AI into various levels of operations CERN’s operations, particularly in the domain of particle accelerator control. The goal is to embed AI assistants in critical areas: aiding control room operations for managing accelerators, assisting in coding for development purposes, and enhancing the effectiveness of documentation and knowledge retrieval. These...
The ATLAS Collaboration is composed of around 6,000 scientists, engineers, developers, students and administrators, with decades of institutional documentation spread across wikis, code docs, meeting agendas, recommendations, publications, tutorials, and project management systems. With the advent of retrieval augmented generation (RAG) and sophisticated large language models (LLMs) such as...
We introduce a summer school workshop designed for a group of gifted students from different backgrounds.
In this workshop, AI language assistants will be employed to aid students in conducting analysis of Open Data from the ATLAS experiment at CERN, with a specific emphasis on the Higgs Boson discovery.
This initiative aims to demonstrate the practical application of AI tools like ChatGPT...
Large language models have - as the name implies - large numbers of parameters. As such not only the training costs but also the inference costs of these models are quite substantial. One strategy for reducing inference costs is to quantize the model weights from 16 bit floating point values to a format with 2-8 bits per weight. However, these custom data formats in turn require custom...
Large language models see rapid adoption in various domains, prompting us to rethink established teaching paradigms. We examine their utility in university-level physics education, focusing on two main aspects: Firstly, how reliable are publicly accessible models in answering exam-style multiple-choice questions? Secondly, how does the question's language affect the models' performance? We...
This work utilizes natural language processing (NLP) techniques to uncover trends and emerging directions in the research about the strong coupling of quantum chromodynamics. We developed an NLP pipeline to extract key topics and trends from abstracts related to strong coupling from the InspireHEP corpus. We performed topic modeling over time which reveals clusters and trends of related ideas...
We report progress in using LLM to generate particle theory Lagrangians. By treating Lagrangians as complex, rule-based constructs similar to linguistic expressions, we employ transformer architectures —proven in language processing tasks— to model and predict Lagrangians. A dedicated dataset, which includes the Standard Model and a variety of its extensions featuring various scalar and...
Recent advances in large language models (LLMs) like chatGPT have demonstrated their potential for generating human-like text and reasoning about topics with natural language. However, applying these advanced LLMs requires significant compute resources and expertise that are out of reach for most academic researchers. To make scientific LLMs more accessible, we have developed Helmholtz...
The surge in observational capabilities and the heightened focus on time-domain astronomy have led to a substantial increase in data volume, reshaping how astrophysicists interpret, process, and categorize information. Despite the utilization of machine-readable data formats in certain instances, a significant portion of information is conveyed through natural language reports. To address the...
In the complex realm of academic research, scholars often grapple with the daunting task of efficiently navigating extensive literature, discerning emerging trends, and evaluating the novelty and feasibility of proposed research ideas. This abstract introduces "MetaInsight," an innovative LLM (Large Language Model)-powered research assistant designed to mitigate these challenges and augment...
Navigating the landscape of particle accelerators has become increasingly challenging with recent surges in contributions. These intricate devices challenge comprehension, even within individual facilities.
To address this, we introduce PACuna, a fine-tuned language model refined through publicly available accelerator resources like conferences, pre-prints, and books.
We automated data...
I will present a multi-modal model that associates astronomical observations imaged by the Hubble Space Telescope with natural language. I will show that the model embodies a meaningful joint representation between the highly-domain-specific images and text using a variety of downstream tasks. The model demonstrates the potential of using generalist rather than task-specific models in parts of...
