IBM has unveiled a significant expansion in its enterprise AI offerings with the introduction of the third generation of its Granite large language models (LLMs). This marks a substantial advancement in IBM's already impressive $2 billion generative AI business, highlighting the company's continued commitment to innovation in this area. The announcement was made today, with the new models poised to bolster IBM's enterprise AI solutions across various fields.
The new Granite 3.0 models come in several configurations, including general-purpose models with 2 billion and 8 billion parameters, as well as Mixture-of-Experts (MoE) variants. Additionally, IBM is introducing models with enhanced guardrail features, termed Granite Guardian models, to ensure safety and effectiveness. These models will be accessible through IBM's watsonX service and other platforms such as Amazon Bedrock, Amazon Sagemaker, and Hugging Face.
Rob Thomas, IBM's senior vice-president and chief commercial officer, expressed his enthusiasm for the pace at which their AI business is growing. He noted during a media briefing that in his 25 years at IBM, he hadn't witnessed a business scaling at such a rapid rate. This growth underscores IBM's strategic emphasis on AI as a cornerstone of its future operations.
The development of Granite 3.0 is not just about scale but also about refinement. Dario Gil, Senior Vice President and Director of IBM Research, explained that the models were trained using a vast dataset of 12 trillion tokens, encompassing both language and code data. This extensive training, combined with architectural innovations, differentiates Granite 3.0 from its predecessors.
Gil also highlighted the performance metrics achieved by the Granite models, claiming superiority over some of the latest models from tech giants like Google and Anthropic. Beyond performance, IBM has prioritised safety, developing advanced "Guardian" models to safeguard against misuse.
Another key aspect of Granite 3.0 is IBM’s commitment to real open source. The models are released under the Apache 2.0 open-source license, which is approved by the Open Source Initiative (OSI). This decision stands in contrast to other models, such as Meta’s Llama, which are not available under an OSI-approved license. The permissive nature of the Apache 2.0 license allows partners to build upon IBM’s technology, fostering a dynamic ecosystem of solutions and applications.
Looking ahead, IBM is setting its sights on what it terms "generative computing", a potential paradigm shift that could redefine how programming is approached. Gil described this as a shift towards programming computers by examples or prompts, rather than detailed, explicit instructions. This concept aligns closely with the capabilities of LLMs like Granite, suggesting a future where generative AI plays an even more integrated role in computing.
IBM's latest announcements underscore its leading position in the enterprise AI market, as the company continues to innovate and expand its offerings to meet growing enterprise demands. This latest suite of Granite models is expected to significantly enhance business processes across customer service, IT automation, BPO, application development, and cybersecurity sectors, further solidifying IBM’s role as a forefront player in the AI space.
Source: Noah Wire Services