本人一直在用ChatGPT、Claude等AIGC工具解决一些日常工作的问题，已经成为我工作“日常”中不可或缺的一部分，也尝试着搭建一些开源的语言模型，这过程让我产生了一些对于未来AIGC迈向何处思考。
近2年,预训练语言模型的进步推动了AI产业的飞速发展,GPT-3等通用语言模型正重塑着我们对人工智能的认知。但是,这仅仅是AI发展道路上的第一个里程碑。真正颠覆商业模式、改造行业的AI浪潮,还需要专业化、细分领域的深度学习能力做支撑。
无论是医疗、法律、教育,还是工业制造等领域,实际业务场景需要的都是融合专业知识的定制AI解决方案,而非万能的通用AI。这就需要训练出理解专业语境、掌握行业知识的AI模型。相比通用模型需要大量数据训练,专业AI模型可以用较少样本进行微调优化,更易于落地实现价值。
与此同时,新兴的生成式AI技术也将成为未来关键。它不仅可以自动生成图像、文本、音频等,还可从少量样本中学习新的任务,无需人工标注大量数据。这类轻量级的生成AI,将助力企业快速应对多变的业务需求。
例如,医学AI助手可以理解专业术语,分析病历和检测结果,对病情变化生成警报,辅助医生制定治疗方案。法律AI可以输入少量法条案例,学习生成法律文书、协议模板,减轻律师repetitive工作量。教育AI可以根据学科知识图谱,自动生成适合不同阶段的教学课件、练习题等。
另一个趋势是轻量化的生成式AI模型。随着移动设备和边缘计算的普及，需要在资源受限的环境中运行AI模型。因此，未来的发展方向将倾向于开发更轻巧、高效的模型，以满足这种需求。这些轻量化的模型可以嵌入到智能手机、智能家居设备等各种终端，为用户提供更便捷、实时的AI支持。
可以预见,伴随模型训练成本的下降,未来将出现更多基于专业领域、适用于具体业务场景的AI应用。而积累多年经验的AIGC,将在连接通用AI和业务现实,推动从通用向专业化演进的过程中,发挥关键作用。
——–以下是chatGPT为此文翻译——
I have been using tools like ChatGPT and Claude from AIGC to solve everyday work-related issues, which have become an indispensable part of my daily routine. I’ve also ventured into building open-source language models, and this journey has led me to ponder about the future direction of AIGC.
Over the past two years, the advancements in pretrained language models have propelled the rapid growth of the AI industry. General language models like GPT-3 are reshaping our understanding of artificial intelligence. However, this is just the first milestone on the path of AI development. The wave of AI that truly disrupts business models and transforms industries requires specialized, domain-specific deep learning capabilities as its foundation.
Whether it’s healthcare, law, education, or industrial manufacturing, practical business scenarios demand customized AI solutions that integrate domain expertise, rather than relying solely on a one-size-fits-all AI. This necessitates training AI models that understand specialized contexts and grasp industry knowledge. Compared to general models that require massive amounts of training data, specialized AI models can be fine-tuned with fewer samples, making them more practical for delivering value.
Simultaneously, emerging generative AI technologies will play a crucial role in the future. These technologies can auto-generate images, text, audio, and more, and can learn new tasks from a small number of samples, eliminating the need for extensive manual data annotation. These lightweight generative AIs will empower businesses to rapidly respond to evolving business needs.
For instance, a medical AI assistant could comprehend professional terminology, analyze medical records and test results, generate alerts for changes in patient conditions, and aid doctors in devising treatment plans. Legal AI could take in a small number of legal cases and statutes, learn to generate legal documents and agreement templates, thus reducing the repetitive workload for lawyers. In education, AI could automatically generate teaching materials and practice questions suitable for different stages based on subject knowledge graphs.
Another trend is the rise of lightweight generative AI models. With the proliferation of mobile devices and edge computing, running AI models in resource-constrained environments is crucial. Therefore, the future development direction leans toward creating more lightweight and efficient models to meet such demands. These lightweight models can be embedded in various terminals like smartphones and smart home devices, providing users with convenient and real-time AI support.
It’s foreseeable that as the cost of model training decreases, more AI applications specific to specialized domains and tailored to specific business scenarios will emerge. Accumulating years of experience, AIGC will play a pivotal role in bridging the gap between general AI and business reality, driving the evolution from general to specialized AI applications.