in

Build Better AI For Enterprise And Hybrid Cloud With IBM’s WatsonX

Build Better AI For Enterprise And Hybrid Cloud With IBM’s WatsonX

AI is hot and IBM put it front and center, along with its hybrid cloud strategy, at its yearly IBM Think conference. While others have focused on the consumer-facing aspects of new AI applications over the last several years, IBM has been developing a new generation of models that will better serve enterprise customers.

IBM announced its AI development platform, called watsonx.ai, for hybrid cloud applications. The IBM watsonx AI development service is in tech preview, with general availability in Q3 of 2023.

IBM BlogIntroducing watsonx: The future of AI for business – IBM Blog

This new generation of AI is designed to be a critical business tool, enabling a new era of productivity, creativity, and value creation. But for an enterprise it’s more than just having cloud access to this new class of AI constructs commonly called Large Language Models (LLMs). LLMs form the basis of Generative AI products such as ChatGPT but, enterprises have many issues that must be factored in: data sovereignty, privacy, security, reliability (no drift), correctness, biases, etc.

An IBM survey of enterprises found that 30-40% are finding business value in AI, which has doubled since 2017. One forecast IBM referenced stated that AI will deliver $16 trillion in global economic contribution by 2030. While this survey calculated productivity improvements using AI, more unique value can be created beyond productivity enhancements, just as no one could have predicted unique future values of the Internet in its early days. AI will fill many gaps between the skill requirements of businesses and people who have those skills by enhancing productivity.

Today AI can already improve software programming by making it faster and more error free. At Red Hat, IBM’s Watson Code Assistant, which uses watsonx, makes it easier to write code by predicting and suggesting the next code segment to enter. This application of AI is very efficient because it targets the specific programming model in Red Hat Ansible Automation Platform. The Ansible Code Assistant is 35x smaller than other, more general code assistants because it’s more bounded and optimized.

Another example is SAP, which will incorporate the Watson service processing to power its digital assistant in SAP Start. New AI capabilities in SAP Start will help boost user productivity with both natural language capabilities and predictive insights using IBM Watson AI solutions. SAP found that up to 94% of queries can be answered by AI.

Bringing watsonx to life

There are three parts of the IBM AI development stack: watsonx.ai, watsonx.data, and watsonx.governance. The watsonx components are designed to work together, and are also open to working with 3rd party integration, such as the open-sourced AI models from HuggingFace. Also, watsonx can run on multiple cloud services, including IBM Cloud, AWS, and Azure, as well as on-premises servers.

The watsonx platform is delivered as a service, and it supports hybrid-cloud deployments. With these tools, data scientists can perform prompt engineering and tuning of custom AI models. The models then become critical engines for enterprise business processes.

The watsonx.data service uses an open table store to allow data from multiple sources to be connected to the rest of watsonx. It manages the life cycle of the data used to train watsonx models.

The watsonx.governance service is used to manage the model life cycle, applying active governance to the models as they are trained and refinedwith new data.

The heart of the offering is watsonx.ai, where the development work takes place. IBM itself has developed 20 foundational models (FMs) today, with different architectures, modalities, and size. On top of those, there’s the HuggingFace open-sourced models that will be available on the watsonx platform. IBM expects some customers will develop applications themselves, but IBM offers consulting to help choose the right models, retrain on customer data, and to help accelerate development when needed.

More than three years of research went into developing the watsonx platform. IBM went so far as to build its own AI supercomputer named “Vela” to research effective system architectures for building FMs (see article link below) and build its own model library before releasing watsonx. IBM served as its own “client 0” for the AI platform.

MORE FROM FORBESIBM’s Cloud AI Supercomputer Vela Builds AI Foundation Models For Enterprise

The Vela architecture is easier and cheaper to build than traditional AI supercomputers using standard Ethernet networking switches (and not using the more expensive Nvidia/Mellanox switches) and is potentially easier for clients to replicate if they want to run watsonx on their premises. PyTorch was also optimized for IBM Vela AI supercomputer architecture. IBM found that there was only a 5% performance overhead to run virtualization on Vela.

IBM’s watsonx supports IBM’s commitment to a hybrid cloud strategy running on Red Hat OpenShift. The watsonx AI development platform runs in the IBM cloud or in other public clouds such as AWS or on customer premises, which allows an enterprise to utilize this latest AI technology even if there are business constraints that do not allow the use of a public AI tool. IBM is truly bringing leading-edge AI and hybrid cloud together with watsonx.

To clarify the naming conventions – watsonx is IBM’s AI development and data platform to deliver AI at scale. The products with the Watson brand name are digital labor products that have an AI expertise. The other Watson branded products are Watson Assistant, Watson Orchestrate, Watson Discovery, and Watson Code Assistant (formerly Project Wisdom). IBM is bringing more focus to the Watson brand. The company has rolled the product previously known as Watson Studio into watsonx.ai, with support for the new foundation model development and access to the traditional machine learning capabilities.

FM and LLMs

Over the last 10 years, deep learning models were trained on massive piles of labeled data for each application. This approach was not scalable. FMs and LLMs are trained on massive piles of unlabeled data, which has become easier to gather. These new underlying foundation models can then used to perform multiple tasks.

The use of the term “LLM” is actually a misnomer for this new class of AI that leverages pretrained models to perform multiple tasks. The use of “language” in the term implies this tech is only suited for test , but models can consist of code, graphics, chemical reactions, etc. The term that IBM uses for these large pretrained models, and which is a more descriptive, is foundation models. With FMs, a large set of data is trained to generate a specific model. This FM can then be used as is, or tuned for a specific application. By tuning the FM for an application, it’s also possible to put in appropriate limits and make the model more useful directly. FMs also can be used to accelerate the pace of non-generative AI applications like data classification and filtering.

Many LLMs are large, and they’re growing larger, because they attempt to train on every type of data so that they can be used for any possible open domain task. In an enterprise setting, that approach is often overkill and may run into scaling issues (see article link below). By properly selecting an appropriate data set and applying it the right type of model, a much more efficient final model can be achieved. This new model can also be cleaned of any bias, copyright material, etc. with IBM’s watsonx.governance.

MORE FROM FORBESGenerative AI Breaks The Data Center: Data Center Infrastructure And Operating Costs Projected To Increase To Over $76 Billion By 2028

Conclusion

At some point during IBM Think, AI was said to be at a “Netscape moment,” an analogy to a watershed moment when a much wider audience was exposed to the capabilities of the Internet. ChatGPT exposed generative AI to a wider audience. But there’s still a need for responsible AI that enterprises can rely on and control.

And as Dario Gil said in his closing keynote: “Don’t outsource your AI strategy to an API call.” That same sentiment was echoed by Hugging face CEO: own your model; don’t rent someone else’s model. IBM is giving enterprises the tools to build responsible and efficient AI, and to own their models.

Tirias Research tracks and consults for companies throughout the electronics ecosystem from semiconductors to systems and sensors to the cloud. Members of the Tirias Research team have consulted for IBM and other companies throughout the server, AI and Quantum ecosystems.

What do you think?

Jim Breyer, Henry Kravis Among the Investors Backing New Digital Platform ‘YourArt’

Jim Breyer, Henry Kravis Among the Investors Backing New Digital Platform ‘YourArt’

Inside Shift Left And API Security From RSA

Inside Shift Left And API Security From RSA