The New York Times and the Daily News encountered an unexpected twist in their copyright lawsuit: an OpenAI engineer inadvertently deleted virtual machine search data that could have been key evidence, adding a dramatic turn to this high-profile legal dispute. According to a letter submitted to the U.S. District Court for the Southern District of New York on Wednesday night, lawyers and technical experts for the two media companies had previously invested over 150 hours searching OpenAI's AI training dataset. However, on November 14, an OpenAI engineer accidentally deleted data stored on the virtual machine.

Recently, research teams from several Chinese scientific institutions have launched the super large-scale multimodal dataset named Infinity-MM, and based on this dataset, they have trained an outstanding AI model called Aquila-VL-2B. This breakthrough injects new momentum into the development of multimodal AI. The Infinity-MM dataset is astonishing in size, containing four major types of data: 10 million image descriptions, 24.4 million general visual instruction data, 6 million selected high-quality instruction data, and 3 million generated by G

A recent study led by the Technion-Israel Institute of Technology reveals that large language models (LLMs) may be more knowledgeable than they appear. Researchers found that LLMs' internal representations encode information about the correctness of their outputs, allowing them to identify the correct answer even when they ultimately generate an incorrect one. The research team focused on analyzing LLM errors in long text generation, which is more aligned with their real-world application scenarios. They constructed an error detection dataset to compare the model's outputs.

In the rapid advancement of artificial intelligence, an international research team is paving the way for the development of AI language models in Europe. They have launched a project called MOSEL (Massive Open-source compliant Speech data for European Languages), creating a comprehensive open-source voice dataset for the 24 official languages of the European Union. This initiative aims to promote the development of open AI language models in Europe, challenging the current dominance of English.

At the 2024 Beijing Cultural Forum, the Beijing Academy of Artificial Intelligence (BAAI) officially announced the release of the next-generation Chinese Internet corpus CCI3.0 (Chinese Corpora Internet), further promoting data co-construction and sharing. CCI3.0 includes a 1000GB dataset and a 498GB high-quality subset CCI3.0-HQ, marking another important update following the initial open-source release of CCI1.0 in November 2023 and the release of CCI2.0 in April 2024.

Recently, OpenAI launched a groundbreaking multilingual dataset designed to evaluate AI performance across 14 languages, including Arabic, German, Swahili, Bengali, and Yoruba. This dataset, named 'Multilingual Massive Multitask Language Understanding' (MMMLU), has been published on the open data platform Hugging Face, marking another significant advancement for OpenAI in the global AI landscape. Dataset link: https://huggingface

Zhiyuan Research Institute has launched a dataset called Infinity-Instruct, which comprises millions of instruction tuning data aimed at enhancing the performance of language models in dialogues and other areas. Recently, Infinity-Instruct has completed a new round of iterations, including the Infinity-Instruct-7M foundational instruction dataset and the Infinity-Instruct-Gen dialogue instruction dataset.

LAION launched Re-LAION-5B, the world's first AI training dataset that fully removes links to CSAM, aimed at addressing the issue of Child Sexual Abuse Material (CSAM). This dataset has been significantly improved over LAION-5B and is mainly divided into two versions: Re-LAION-5B Research and Research-Safe. A total of 2,236 CSAM links have been removed, including 1,008 from child protection organizations' lists. The dataset contains 5.5 billion pairs of text and images, designed to help

The German research institution LAION has created and released a new dataset, Re-LAION-5B, claiming to have removed links to suspected Child Sexual Abuse Material (CSAM). This dataset is a revised version of the old LAION-5B dataset, incorporating recommendations from the Internet Observation Foundation, Human Rights Watch, the Canadian Centre for Child Protection, and Stanford Internet Observatory. Re-LAION-5B is divided into two versions: Research and Research-Safe, with the latter additionally removing NSFW content. The dataset aims to support

As AI technology rapidly advances, the challenges of data acquisition are becoming increasingly prominent. Research from institutions like MIT reveals that open-source datasets such as C4, RefineWeb, and Dolma are facing stricter licensing agreements from the websites they scrape, posing significant challenges for AI training and academic research. The study found that even major AI companies like OpenAI are subject to strict limitations due to r....

"MedTrinity-25M" is a large-scale multimodal medical dataset released by the UCSC-VLAA team, containing 25 million medical images along with detailed annotations. This dataset features multi-granularity annotations, supports the training of multimodal large medical models, and involves a complex construction process that includes data processing, metadata extraction, region of interest localization, and the collection of medical knowledge. Detailed descriptions are generated using large-scale language models to enhance data usability. The dataset was officially released on July 21, 2024, accompanied by a pre-trained model.

Recently, the 'Diting' seismic wave large model, jointly developed by the National Supercomputing Center in Chengdu, the Institute of Geophysics of the China Earthquake Administration, and Tsinghua University, was officially released in Chengdu, Sichuan. This model is the first seismic wave large model in the country to reach 100 million parameters, marking a significant breakthrough in the integration of seismology research and artificial intelligence technology in China.

Wuhan University, in collaboration with China Mobile's Jiutian AI team and Duke Kunshan University, has released the open-source audio-video speaker recognition dataset VoxBlink2, which is based on YouTube data and contains over 110,000 hours of audio-video recordings. The dataset includes 9,904,382 high-quality audio clips and their corresponding video segments, sourced from 111,284 users on YouTube, making it the largest publicly available audio-video speaker recognition dataset to date. The release of this dataset aims to enrich open-source speech corpora and support the training of voiceprint large models.

Recently, researchers at Microsoft have introduced a novel AI framework called Auto Evol-Instruct, which can automatically evolve instructional datasets without any human intervention.In the field of artificial intelligence, the development of large language models (LLMs) is crucial, especially in enhancing their ability to follow detailed instructions. Researchers have been exploring ways to improve the datasets used to train LLMs to enhance their performance and adaptability.Traditional evolut

Recently, an investigation revealed that several tech giants, including Apple, had utilized YouTube video subtitles to train AI models. These data encompassed over 170,000 videos, including content from well-known creators such as MKBHD and Mr. Beast. Apple used this data to train its open-source model, OpenELM, which was released in April of this year.In response, Apple recently clarified externally that OpenELM has not been applied to any of its AI or machine learning functions, including Appl

A French startup named Presti is revolutionizing product photography in the furniture industry by leveraging generative artificial intelligence (GenAI). Traditionally, furniture brands have incurred substantial costs for photo shoots when launching new collections, a process that is not only cumbersome but also expensive. Presti aims to address this issue by transforming single product images into realistic lifestyle photos.Image source: The image is AI-generated, provided by the image licensing

IBM opened the Granite13B LLM model for enterprise applications in May. Now, Armand Ruiz, the vice president of IBM's AI platform products, has publicly disclosed the full content of the comprehensive 6.48TB dataset used to train Granite13B.The dataset, after strict preprocessing, was reduced to 2.07TB, a reduction of 68%. Ruiz emphasizes that this step is crucial for ensuring a high-quality, unbiased, ethical, and legally compliant dataset to meet the needs of enterprise applications. The datas

Nvidia recently announced the launch of two groundbreaking AI technologies: HelpSteer2 and Llama3-70B-SteerLM-RM. These innovations promise to significantly enhance the capabilities of AI systems across a variety of applications, from autonomous driving to natural language processing.