Questa è la nostra ultima selezione di pubblicazioni e brevetti in inglese su Large Language Models (LLM), tra numerose riviste scientifiche online, classificate e incentrate su large language model, LLM, generative pre-trained transformer, pre-training, transformer architecture, gradient descent, GPT, tokenization, generative model, self-attention mechanism, masked language model e MLM.
Improvements to language model watermarking
Patent published on the 2025-03-06 in WO under Ref WO2025050084 by VERANCE CORP [US] (Winograd Joseph [us])
Abstract: A method of embedding a watermark into an Al LLM output comprising inputting a text prompt to an LLM, the LLM generating a first token text output. Binary reduction is performed on the token text output and generating watermarks using a watermark-seeded random number generator. The watermarks are applied to successive bits of the character set representations of the output by making a decision about accepting a completed token using a second random number generator that is distinct from the firs[...]
Our summary: Method of embedding watermarks into an AI language model output by performing binary reduction on token text output and applying watermarks using a watermark-seeded random number generator with acceptance decision based on relative token probability. New model PDF generated for accepted tokens.
language model, watermarking, binary reduction, random number generator
Patent
Agentic based processing for carbon emissions auditing
Patent published on the 2025-03-06 in WO under Ref WO2025049513 by GEOQUEST SYSTEMS B V [NL] (Manikani Sunil [in], Sule Vishwanath [in], Hassan Abbi Moghaiyera [us], Milne Paul [gb])
Abstract: Carbon emission auditing includes obtaining a supplier transaction record of an enterprise corresponding to a supplier entity from a transaction repository. A research response corresponding to the supplier entity is obtained from a large language model (LLM). A validity of the supplier transaction record based on the research response is further obtained from the LLM as a validation response. Field values of the record fields of the supplier transaction record are further verified by the LLM, a[...]
Our summary: Agentic based processing for carbon emissions auditing involves obtaining supplier transaction records, validating them using a large language model, and generating audit failure explanations.
agentic, carbon emissions, auditing, language model
Patent
Techniques for generating musical plan based on both explicit user parameter adjustments and automated parameter adjustments based on conversational i
Patent published on the 2025-03-06 in WO under Ref WO2025049565 by AIMI INC [US] (Balassanian Edward [us], Sorensen Andrew C [us], Hutchings Patrick E [us])
Abstract: Disclosed techniques relate to user control of generative music. In some embodiments, a computing system generates a musical plan based on both conversational inputs (e.g., using a large-language model (LLM)) and non-conversational inputs (e.g., via a traditional user interface) to a hybrid interface. The computing system may generate an initial version of the musical plan based on the LLM context and update the context and plan based on various types of user input via the hybrid interface. Disc[...]
Our summary: Techniques for generating musical plan using user and automated parameter adjustments, including conversational inputs and traditional user interface, allowing guided user control over generative music systems.
generative music, user control, computing system, hybrid interface
Patent
Methods and systems for synthesizing raw multichannel data
Patent published on the 2025-03-06 in WO under Ref WO2025049719 by THE REGENTS OF THE UNIV OF CALIFORNIA [US] (Deveshwar Nikhil [us], Rajagopal Abhejit [us], Larson Peder [us])
Abstract: Methods of synthesizing raw multichannel data are provided. The subject methods include receiving initial magnitude image data and producing phase image data from the initial magnitude image data using a generative model to synthesize the raw multichannel data. Systems and computer-readable media for practicing methods of the invention are also provided.[...]
Our summary: Methods and systems for synthesizing raw multichannel data, including receiving initial magnitude image data and producing phase image data using a generative model. Systems and computer-readable media for practicing these methods are also provided.
multichannel data, synthesizing, generative model, computer-readable media
Patent
Automated inventory management system
Patent published on the 2025-03-06 in WO under Ref WO2025050097 by STAXEL INC [US] (Dobrynin Dmitrii [us], Zavadskiy Victor [us])
Abstract: Systems and methods for automated warehouse fulfillment. Modular storage units may be installed in warehouse racks. The storage units may comprise rows of cell compartments configured to hold modular product cells. Each product cell may hold a single product and each of the product cells may be interconnected with one another to form chains of cells. These chains of cells may be loaded in the cell. Grabbing assemblies on automated robots may be used to load, unload and move product cells. One or[...]
Our summary: Automated inventory management system using modular storage units in warehouse racks, with automated robots for loading and unloading product cells to fulfill orders efficiently.
automated inventory management, warehouse fulfillment, modular storage units, automated robots
Patent
Systems and methods of managing an interaction with a large language model
Patent published on the 2025-03-06 in WO under Ref WO2025050056 by MARLIN MARISA [US] (Marlin Todd [us], Marlin Marisa [us])
Abstract: The present disclosure provides a method of managing an interaction with a large language model (LLM). Further, the method may include receiving a prompt data from a user device associated with an organization. Further, the method may include receiving a policy data from an organization device associated with the organization. Further, the policy data may be associated with the organization. Further, the method may include analyzing the prompt data based on the policy data. Further, the method m[...]
Our summary: Systems and methods for managing interactions with a large language model by analyzing prompt data based on policy data, generating output data using the language model, and storing and transmitting relevant information.
language model, interaction management, policy data, output data
Patent
Improving Probability-based Prompt Selection Through Unified Evaluation and Analysis
Published on 2024-05-25 by Sohee Yang, Jonghyeon Kim, Joel Jang, Seonghyeon Ye, Hyunji Lee, Minjoon Seo @MIT
Abstract: Previous works in prompt engineering for large language models have introduced different gradient-free probability-based prompt selection methods that aim to choose the optimal prompt among the candidates for a given task but have failed to provide a comprehensive and fair comparison between each other. In this paper, we propose a unified framework to interpret and evaluate the existing probability-based prompt selection methods by performing extensive experiments on 13 common and diverse NLP ta[...]
Our summary: Evaluation of probability-based prompt selection methods through unified framework, Improving prompt selection effectiveness through combinatorial variants of mutual information, Introducing Calibration by Marginalization method for unbiased prompt selection, Achieving high performance in prompt selection without calibration by maximizing mutual information.
prompt selection, probability-based, unified evaluation, analysis, NLP tasks
