7 Take-Aways from Alibaba Cloud’s AI and Big Data Summit (Tech Strategy – Daily Article)

Posted on by

Alibaba Cloud recently held their “AI and Big Data Summit” in Singapore. It was pretty good. And it had pretty much everyone I wanted to hear from:

  • Selina Yuan – Vice President of Alibaba Group; President of Alibaba Cloud International
  • Zhou Jingren – Chief Technology Officer for Alibaba Cloud
  • Junhua Wang – Vice President, Alibaba Cloud; Head of Computing Platform, Alibaba Cloud
  • Dongliang Guo – Vice President, Product and Solution, Alibaba Cloud International
  • Chris Tung – President of Alibaba Group, Strategic Development

That’s pretty much my interview wish list for this company.

This article is my list of 7 take-aways. Basically, my notes on what I thought was important.

#1 – Alibaba Cloud Now Describes Itself as the Digital Technology AND INTELLIGENCE Backbone of Alibaba Group.

The addition of the word “intelligence” is notable. Building intelligence capabilities into companies is the next big thing. It’s the big topic I’m going to focus on over the next five years.

Alibaba Cloud is currently the leading cloud player in China. And it’s in the top tier internationally. Particularly in Infrastructure as a Service. Their cloud services are what you would expect. Here’s how they describe them:

  • “Cloud services include elastic computing, database, storage, network virtualization services, largescale computing, security, management and application services, big data analytics, a machine learning platform and IoT services.”

But Alibaba Cloud is also one of the leading AI and generative AI companies for China. They are on the short list of majors to watch (Baidu and Tencent Cloud are others. Maybe JD). But being a cloud service company puts them in a great position to see what is actually being adopted across industries.

They are also putting LLMs into all of the Alibaba products. It appears to be going into absolutely everything. Here’s the list.

#2 – Efficient and Scalable Computing Is the Current GenAI Bottleneck

There is an increasing demand for generative AI technologies. Like from virtually every company. So, we have seen a big fight for GPUs. Which has been fantastic for Nvidia.

But the bigger problem is the huge compute power needed for GenAI in general. Foundation models just requires a ton of computing. This creates an expensive step for clients. And then there is the problem of scaling up efficiently. As models and usage grows, it’s easy for the costs to go crazy. Efficient and scalable computing is a problem in GenAI.

That is good for cloud services (like Alibaba Cloud) as they offer scalable services and can spread the workload across cloud-based servers. The speakers at the conference talked a lot about how cloud services (i.e., serverless) can use GPUs efficiently with high-speed networking. And how they are working on how to scale training and inference without losing efficiency.

Stability was mentioned as another challenge. Foundation models and GenAI apps are huge systems. And the bigger the system, the more likely there will be errors. It’s a problem training a model for weeks and months – and you’re getting errors every couple of hours. So, stability is a big priority as you scale up. The goal is to run tons of GPUs for weeks without interruption.

#3 – Efficient Data Processing Is the Next GenAI Bottleneck

In addition to computing power, GenAI also requires a huge amount of data for training and ongoing inference. Processing the data flowing in and out of the LLM is a big undertaking. There is tons of data cleaning, tagging, removing duplication and so on. And this is mostly unstructured data.

The speakers talked a lot about the need to process unstructured data efficiently. At large scale. How can you speed this up? How do you unify data preparation and training?

They also mentioned that inference is much harder than training. This is the ongoing data flows that support usage of the models. They described inference data processing as “the new battleground”.

At the conference, they introduced MaxCompute (previously known as ODPS), which is “a general purpose, fully managed, multi-tenancy data processing platform for large-scale data warehousing.” Basically, it’s a big upgrade to their big data service.

They are rolling out products for the two bottlenecks just described (compute and data processing).

#4 – Alibaba Cloud’s Primary Focus is Making Model Training Easier with “Model as a Service”.

Alibaba Cloud’s approach to generative AI is to open source their language models and offer a suite of solutions for developing customized generative AI applications.

That’s basically Model as a Service. And it’s about making training, customizing, and deploying LLMs easier. And cloud is a natural home for this. Cloud services are already pay as you go, ready to use, and have automatic scaling.

Here’s their service suite.

There’s Infrastructure as a Service (IaaS). That’s computing, storage, networking, security, etc.

There’s Platform as a Service (PaaS). That’s databases, big data analytics, data warehouse and data lake. Also, middleware and container service.

And then there’s Model as a Service (MaaS). This includes:

  • Training
  • Inference
  • Foundation models
  • Development tools and community
  • A growing suite of AI apps – some developed in-house but most by developers. Such as:
    • Intelligent Voice Interactions
    • NLP
    • Video Intelligence
    • Intelligent customer service
    • Industry intelligence
    • Decision intelligence

#5 – Qianwen is the Engine of Alibaba’s GenAI Strategy

Tongyi Qianwen is Alibaba’s response of OpenAI’s GPT, Baidu’s Wenxin and Google’s LaMDA. This is their rapidly growing family of open-source AI foundation models. All their services are built around deploying and customizing them.

However, none of their suite of GenAI services are foundation model specific. You can use most all the large 3rd party foundation models (both open source and commercial). This agnostic approach mirrors the strategy taken by Baidu and Huawei Cloud.

Today, Tongyi Qianwen (Qwen for short) has a family of 4 LLMs and 2 multimodal models.

Interestingly, Qwen is multilingual. It has Chinese and English capabilities. And it also incorporates Southeast Asian, East Asian, and Middle Eastern languages. That is an interesting differentiation, which somewhat mirrors how Chinese and South Korean search engines are different than Google Search.

An example presented at the conference was by rinna, a Japanese startup specializing in the development of pre-trained foundation models in Japanese. They have launched several “Nekomata models”, which are based on the open-source Tongyi Qianwen LLMs, namely the Qwen-7B and Qwen14B. A presenter from rinna mentioned that the Qwen vocabulary significantly enhanced the Nekomata models’ ability in Japanese text compared to their earlier series which were based on the Llama2 architecture.

#6 – Qwen is Complemented by ModelScope and Model Studio

The adoption and usage of both their services and foundation models is the key. Those are the KPIs to watch for Alibaba Cloud in GenAI. That means providing lots of support to both customers and technology partners, especially developers. The goal is to facilitate adoption and integration into business scenarios and applications.

So, they have launched ModelScope, an AI model community and library. This is similar to HuggingFace, the library and open-source hub for AI experts.

They also have Service Platforms, specifically AI Cloud Model Studio and PAI Dashscope (their API Service). Model Studio has Alibaba Cloud’s customization and application services. This is where customers can integrate their own data, do their training, do fine tuning, inference and other steps for deployment and customization.

#7 – Developers and Technology Partners Will Determine the Winners

These are the early days of generative AI. The new AI tech stack is taking shape. And cloud companies are positioned to win with innovation platform business models. But it will come down to which platforms and technologies the developers and companies decide to build on. There needs to be lots of companies and developers building on their system and with their tools. It’s a race for participation. And only a small number of models are going to survive. If developers aren’t writing apps that run on Qwen, then it will fade away.

Alibaba Cloud is 100% pitching hard to developers and technology partners. And it looks like they have the winning cards. In their offerings, you can hear a very clear pitch to developers and technology partners.

  • Join the Alibaba Cloud Technology Partner Program.
  • You get access to +4M customers.
  • You will get joint promotion.
  • We provide services for easy landing and adoption.
  • We are focused on being open source and technology agnostics.
  • We are supporting lots of GitHub projects.
  • We have a cloud startup program.
  • We have an academy and classes in generative AI.

And so on. We’ll see.

***

Those are my notes from the summit. The big questions going forward are:

  • Who is getting the most adoption? By companies? By developers and technology partners?
  • What are the hot use cases?

That’s it. Cheers, Jeff

———–

Related articles:

From the Concept Library, concepts for this article are:

  • Cloud Services
  • AI: Generative AI

From the Company Library, companies for this article are:

  • Alibaba Cloud

Photo of Selina Yuan, President of Alibaba Cloud (press release)

———-

I write, speak and consult about how to win (and not lose) in digital strategy and transformation.

I am the founder of TechMoat Consulting, a boutique consulting firm that helps retailers, brands, and technology companies exploit digital change to grow faster, innovate better and build digital moats. Get in touch here.

My book series Moats and Marathons is one-of-a-kind framework for building and measuring competitive advantages in digital businesses.

This content (articles, podcasts, website info) is not investment, legal or tax advice. The information and opinions from me and any guests may be incorrect. The numbers and information may be wrong. The views expressed may no longer be relevant or accurate. This is not investment advice. Investing is risky. Do your own research.

Leave a Reply