IBM CEO praises real open source for enterprise gen AI, new efforts emerge at Think 2024
[ad_1]
Join us in returning to NYC on June 5th to collaborate with executive leaders in exploring comprehensive methods for auditing AI models regarding bias, performance, and ethical compliance across diverse organizations. Find out how you can attend here.
IBM is strengthening the foundation for its generative AI efforts with a new set of technologies and partnerships being announced today at the company’s Think 2024 conference.
IBM has a long history in AI that predates the modern hype era of gen AI by decades. It was however only a year ago at Think 2023, that IBM announced its Watsonx genAI product platform. The Watsonx platform has become the foundation of IBM’s gen AI efforts, providing organizations with enterprise-grade models, governance and tools. At Think 2024, IBM is making a series of its Granite models fully available as open-source code. The Granite models range from 3 to 34 billion parameters and include models for both code and language tasks. Beyond its own models, IBM will be bringing Mistral AI models to its platform, including Mistral Large.
Among the most common use cases for gen AI is for building assistants. It’s a use case IBM is looking to further support with new Watsonx assistants.
“AI is going to be a massive opportunity,” IBM CEO Arvind Krishna said during a roundtable with press. “I actually believe that AI’s impact is going to be on the same order as the steam engine or the internet.”
VB Event
The AI Impact Tour: The AI Audit
Request an invite
Why open source is the foundation for IBM Granite enterprise AI
IBM first announced its Granite models in September of 2023 and it has been expanding the models ever since.
Among the models is a 20 billion parameter base code model that helps to power IBM’s watsonx code assistant for Z service that can be used by organizations to modernize old COBOL applications.
What’s new at Think 2024 is IBM officially making a group of its most advanced Granite models available under the open-source Apache license. It’s important to note that while many vendors claim to have open models, very few are in fact licensed under an Open Source Initiative (OSI) approved open source license. In fact, IBM claims that only Mistral and now Granite are the only highly performant large language model (LLM) model families available under a bona fide open-source license, such as Apache.
In response to a question from VentureBeat, Krishna strongly emphasized that real open source matters for reasons that enterprises should care about. Krishna said that unlike the Apache license which is a real open-source license, there are a host of other open licenses that vendors use that aren’t really open source.
“They’re using the term open just for marketing purposes,” Krishna said.
Krishna explained that real open source is critical to enable contributions and growth for a technology.
“If you want people to contribute, it has to be clearly open source, it cannot actually just be open source marketing,” he said.
IBM expanding WatsonX assistants to advance enterprise AI
While LLMs are of course critical to enterprise gen AI, so too is the concept of an AI assistant.
AI assistants, which some companies like Microsoft and Salesforce call a copilot, is a more consumable approach for gen AI for many types of organizations. During the media roundtable briefing, Rob Thomas, senior vice president and Chief Commercial Officer at IBM explained that the AI assistant provides more of a packaged approach to how an enterprise can put AI into production.
At Think 2024 IBM is announcing three new assistants. The first is Watsonx code assistant for Java, which helps developers to write Java application code. IBM is using its decades of experience with Java to help inform the model.
The second new AI assistant is Watsonx assistant for Z. The IBM Z is IBM’s mainframe system architecture. Thomas explained that the WatsonX assistant for Z is all about helping organizations to manage IBM Z environments. The third new service is the Watsonx Orchestrate, which Thomas said enables enterprises to build their own assistants.
RAG, vector databases and InstructLab
Among the most common enterprise deployment patterns for gen AI today is Retrieval Augmented Generation (RAG).
RAG can help to empower assistants and gen AI chatbots with real enterprise information that an LLM was not originally trained to know. At the foundation of RAG is some form of vector database, or vector support within an existing database. While RAG and vector databases are critical aspects of modern enterprise AI, IBM is not building its own vector database.
“In terms of vector databases, this is kind of like a flavor of the month thing, meaning I think there’s hundreds of options,” Thomas said. “We’ve done integrations with many and by definition, you have to have the capability in the platform, but we don’t feel like we need to own that capability.”
While RAG is important, IBM sees a real future for the InstructLab technology that IBM’s Red Hat business unit recently announced. Dario noted that InstructLab enables continuous improvement of models in an optimized approach.
Will gen AI create jobs or destroy them?
Like most of the enterprise IT industry, IBM is extremely bullish on the prospects of gen AI.
IBM is also very aware of the potential impact it can have on society and employment. During the roundtable, Krishna was asked if there is an employment issue with gen AI and he provided a very succinct response.
“People who are more productive will usually get more business, so the total employment will increase as total productivity increases,” Krishna said. “Except for a couple of countries, demographics are pointing towards having fewer and fewer people, so actually, unless we get these capabilities, we’re not going to be able to grow our quality of life and underlying GDP.”
[ad_2]
Source link