Not to be outdone in the generative artificial intelligence (AI) race, China-based technology giant Alibaba recently announced a number of its own tools and models.
Its flagship product is ChatGPT’s challenging large-scale language model (LLM) called Tongyi Qianwen. It has been cited as one of the largest and most powerful generated chatbots ever created.
Alibaba is best known in the West as an e-commerce giant, similar to Amazon in many ways. Like its American company, the company is also a world leader in cloud services offered through its subsidiaries. alibaba cloud.
Our dominance across retail and web services means we are well-positioned to create apps that enable enterprise customers to start their own generative AI experiments, unlocking their transformative power, driving sales, and You can create a new type of customer experience.
So let’s take a look at some of the ways the company does this and how its efforts differ from those of its competitors on both sides of the globe.
The truth from a thousand questions
Although there is no official English name at this time, this is a close translation of the Chinese name “Tongyi Qianwen”. The name refers to the fact that this, like ChatGPT, is positioned first as a question-and-answer bot.
However, like its competitors, it can also act as your coding co-pilot, generating code in numerous computer programming languages and automatically error checking.
Use cases identified so far include integration with Taobao, the company’s retail business, to create a service known as Taobao Ask. the current, beta testThe service allows users to ask for customized product recommendations and compare different products and services. Users who accessed the test say it can provide improved recommendations compared to traditional search methods.
It’s also integrated into Tingqu, the voice AI that powers smart speaker devices like the Echo. Tmall Genie, and its workplace messaging app, DingTalk. Here you can draft meeting summaries, create personalized schedules, create copywriting for marketing purposes, and generate action plans and to-do lists based on sketches and handwritten notes. can.
As Zhou Jingren, CTO of Alibaba Cloud, state“As we gradually integrate Tongyi Qianwen models into our products and services, we hope that our users will be able to benefit from these fascinating AI innovations in their work, study, play and interaction.”
Beyond this, Alibaba’s strategy appears to be focused on enabling companies to develop their own generative AI solutions built around Tongyi Qianwen and other generative models.To achieve this goal, the company launched model scope, a Model-as-a-Service platform that provides access to hundreds of pre-trained generation tools through a cloud platform. The company has established partnerships based on technology deployment with smartphone manufacturer Oppo and Zhejiang University.
Another major use case for generative AI is image generation, and Alibaba is riding on this trend thanks to its Tongyi Wanxiang model. It is based on a diffusion technique similar to that used in . stable diffusion or The middle of a journeya computer learns how to construct an image by adding “noise” to the data, and then learns how to reconstruct an image by reversing that process.
Next, we use a natural language model (Tongyi Qianwen) to insert text prompts with instructions about what the new image should look like. The Wanxiang model builds on existing diffusion techniques by integrating new models developed by Alibaba researchers. Compositiongives you more control over the model’s output.
open source questions
Unlike competing platforms such as ChatGPT and Google Bard, Alibaba has chosen to open source many of the models available through Tongyi Qianwen and Modelscope.
This means that the source code is public and anyone can freely modify it. Open sourcing a technology has several benefits for its creators, as it means they can incorporate improvements, changes, or re-engineering made by developers around the world.
Instead, it means anyone can peek inside and see exactly how it works. This is not the case, for example, with other models such as OpenAI’s GPT-4, which remains a black box.
However, part of the reason for choosing this path may have been to remain compliant with China’s recently enacted AI law.This law is known as Interim measures regarding generative artificial intelligence service managementcalls on providers of publicly available AI services to “take effective steps to improve the transparency of their generated AI services.”
However, the company is not alone in choosing the open source path. Meta (formerly Facebook) did something similar with his own Llama-2 LLM released this year.Only after the source code leaked on the internet. To further demonstrate how closely connected all the work being done on LLM and generative AI is, Alibaba also recently announced a model based on Llama-2, known as Llama-2. megatron llama.
However, Alibaba has announced that it will open source two models for its LLM, Qwen-7B and Qwen-7B-Chat, both pre-trained with 7 billion parameters.
Alibaba CEO Eddie Wu said the company’s two core priorities going forward will be to become more customer-centric and more AI-driven.
This showed that it was fully ready to actively participate in the “100 Model War”. semester A term coined by Tencent Vice President Jiang Jie to describe the current rush to build and deploy generative AI in Chinese industry.