Taking on Large Language Models: BigScience Workshop Tackles Innovation and Research for the 2022 Language AI Market

采用大型语言模型:BigScience Workshop致力于2022年语言AI市场的创新和研究

2022-01-19 09:00 CSOFT

本文共459个字,阅读需5分钟

阅读模式 切换至中文

As we have explored extensively in our recent posts on advancing language AI, the most dynamic developments in areas like NLG and NLP have tended to come via large language models (LLMs) that require massive and high-quality datasets to train, which naturally favors large companies and software groups who alone manage access to these datasets. However, as the AI market continues to evolve, tech companies of all sizes continue to have a vested interest in enhancing language AI through growth that is largely propelled by the advanced research and data. Now, following the developments from an international collective research project dubbed BigScience, data scientists and researchers have collaborated with the shared aim to promote innovation throughout the broader AI community and help companies to better-address the leading challenges in the field of language AI. Beginning in May of 2021, BigScience is a yearlong research workshop involving the collaboration of 600 researchers from multiple disciplines, representing nearly fifty nations in a collective effort to develop a massive neural network and multilingual text dataset, all on a supercomputer provided by the French government. While presented as a grassroots approach to commercializing LLMs and open-source language software, BigScience aims to not only address significant barriers to entry in the field of language AI, but to also develop new ways of utilizing datasets and new deployments for language AI technology. Specifically, finding solutions to some of the leading technological shortcomings – mainly language coverage in text datasets, and improving accessibility to training models – is one of the more important focusses in this project. Moreover, BigScience aims to explore questions like the environmental and social impacts of LLMs and supercomputers, specialized areas that require a multidisciplinary approach to research. For smaller-tech companies, especially, BigScience is enabling developments in a field that is often associated with tech giants with seemingly endless resources. With innovative approaches to challenges throughout the industry, projects like BigScience extend well beyond the AI community and into the very businesses that deploy AI products. As consumer preference for functionalities like chatbots becomes prevalent across global markets and languages, demand increasingly exceeds supply for high quality training data in those languages and expanding access developers in less dominant areas of the industry bodes well for improving this. Ultimately, the significance of BigScience is that it is helping drive innovation away from experimental, out-of-reach AI phenomena and toward consumer-facing products and applications that will increasingly drive the flow of technologies and services across borders. As language AI improves, translation – one of its vanguard real-world uses – will only become more important to ensuring these technologies succeed with global users. To learn more about CSOFT’s cutting-edge language translation services and customized localization solutions, visit us at csoftintl.com!
正如我们在最近关于推进语言AI的文章中广泛探索的那样,在NLG和NLP等领域中最具活力的发展往往来自大型语言模型,这些模型需要大量高质量的数据集来训练,这自然有利于那些单独管理这些数据集访问的大公司和软件集团。然而,随着人工智能市场的不断发展,各种规模的科技公司在通过主要由先进的研究和数据推动的增长来增强语言人工智能方面继续拥有既得利益。现在,随着一个名为“BigScience”的国际集体研究项目的进展,数据科学家和研究人员携手合作,共同致力于推动整个人工智能领域的创新,并帮助企业更好地应对语言人工智能领域的主要挑战。 从2021年5月开始,BigScience是一个为期一年的研究研讨会,涉及来自多个学科的600名研究人员,代表近50个国家,共同努力开发一个大规模神经网络和多语言文本数据集,所有这些都是在法国政府提供的超级计算机上进行的。虽然BigScience是将LLMs和开源语言软件商业化的草根方法,但它的目标不仅是解决语言AI领域的重大进入壁垒,而且还开发利用数据集的新方法和语言AI技术的新部署。具体地说,为一些主要的技术缺陷找到解决方案--主要是文本数据集中的语言覆盖,以及改进训练模型的可访问性--是本项目中比较重要的焦点之一。此外,BigScience的目标是探索像LLM和超级计算机的环境和社会影响这样的问题,这些专业领域需要多学科的研究方法。尤其是对小型科技公司而言,BigScience正在推动一个领域的发展,而这个领域往往与拥有似乎无穷无尽资源的科技巨头联系在一起。 BigScience等项目以创新的方式应对整个行业的挑战,远远超出了人工智能社区,深入到部署人工智能产品的企业。随着消费者对聊天机器人等功能的偏好在全球市场和语言中变得普遍,对这些语言的高质量训练数据的需求日益超过供应,而在行业不太主导的领域扩大访问开发人员,预示着改善这一状况的好兆头。归根结底,BigScience的意义在于,它正在帮助推动创新从实验性的,遥不可及的人工智能现象转向面向消费者的产品和应用,这些产品和应用将越来越多地推动技术和服务的跨国流动。随着语言AI的发展,翻译--它最重要的现实用途之一--对于确保这些技术在全球用户中获得成功只会变得更加重要。 欲了解更多关于Csoft的尖端语言翻译服务和定制化本地化解决方案,请访问Csoftintl.com!

以上中文文本为机器翻译,存在不同程度偏差和错误,请理解并参考英文原文阅读。

阅读原文