Self Hosted LLMs for secure, trustworthy and multilingual customer service AI

这是Algomo（英国）和Gigit（新加坡）之间的合作项目。深度学习的进步使大型语言模型（LLM）成为客户服务以及其他商业应用的潜在游戏规则改变者。这些模型可以在没有明确指令或手动数据的情况下生成自然的交互，不像当前的会话式人工智能（cAI）系统。然而，这些模型是众所周知的庞大和昂贵的训练和服务，因此，只有大型科技巨头（如谷歌，Meta和OpenAI）有资源来训练，维护和服务它们。任何想要利用LLM潜力的公司，只能通过直接API访问它们。然而，这为客户服务应用程序带来了以下问题：1. ** 语言多样性 **：大多数LLM都是在英语数据上训练的，而在东南亚和非洲的许多国家，英语不是主要语言，这使得这些模型对这些地区实际上毫无用处，并剥夺了他们各自人口的很大一部分。** 隐私问题：** LLM已经接受了关于公共数据集的培训，这些数据集包含PII信息，通过生成PII数据输出构成重大网络安全威胁。此外，直接访问API意味着公司需要在其场所之外共享私人信息，这引发了道德，监管和运营方面的担忧，特别是对于银行，保险和医疗保健等隐私敏感行业。为了解决这些问题，Algomo，一家致力于客户服务多语言AI的英国公司，将与Gigit合作，一家位于新加坡的公司，致力于自主托管和隐私保护的LLMS。在这个项目中，我们将开发一个自主托管的解决方案和工作流程，使培训和托管多语言LLM，这些解决方案对于将我们的多语言对话AI带给非技术用户至关重要，并有助于我们将AI的进步带给地球上的每一个人的使命，无论他们说什么语言。

This is a collaborative project between Algomo (UK) and Gigit(Singapore).Advances in Deep Learning have made Large Language Models (LLMs) a potential game-changer for customer service, as well as other commercial applications. These models can generate natural interactions without explicit instructions or manual data, unlike current conversational Artificial Intelligence (cAI) systems.However, these models, are notoriously large and expensive to train and serve, and as such, only large tech giants (such as Google, Meta and OpenAI) have the resources to train, maintain and serve them. Any company that wants to harness the potential of LLMs, can only access them through direct APIs.However, this poses the below issues for customer service applications:1. **Linguistic diversity**: Most LLMs are trained on mostly English data while there are many countries in Southeast Asia and Africa where English is not the main language, making these models practically useless for these geographies, and disenfranchising a significant part of their respective populations.2. **Privacy issues:** LLMs have been trained on public datasets, which contain PII information, posing a significant cybersecurity threat by generating PII data outputs. Moreover, direct API access means that companies need to share private information outside their premises, which raises ethical, regulatory and operational concerns, especially for privacy-sensitive industries like banking, insurance, and healthcare.To address these issues, Algomo, a UK company working on multilingual AI for customer service, will collaborate with Gigit, a company based in Singapore working on self-hosted and privacy-preserving LLMS.In this project, we will develop a self-hosted solution and workflow enabling training and hosting a multilingual LLM, that can be directly integrated with Algomo.These solutions are essential to bringing our multilingual conversational AI to non-technical users and contribute to our mission of bringing the advancements of AI to every person on earth, no matter what language they speak.