Can AI designed for the enterprise (for example, AI that automatically completes reports, spreadsheet formulas, and so on) ever be interoperable? Along with a number of organizations including Cloudera and Intel, the Linux Foundation—the nonprofit that supports and maintains a growing number of open source efforts—aims to find out.
The Linux Foundation on Tuesday was announced the launch of the Open Platform for Enterprise AI (OPEA), a project to promote the development of open, multi-provider and composable (i.e. modular) productive AI systems. Under the purview of LFAI and the Linux Foundation’s Data org, which focuses on AI and data-related platform initiatives, OPEA will aim to pave the way for the release of “hardened”, “scalable” genetic AI systems that “leverage the best open source innovation from across the ecosystem,” LFAI and Data executive director Ibrahim Haddad said in a press release.
“OPEA will unlock new possibilities in artificial intelligence by creating a detailed, synthetic framework that sits at the forefront of technology stacks,” said Haddad. “This initiative is a testament to our mission to foster open source innovation and collaboration within the AI and data communities under a neutral and open governance model.”
In addition to Cloudera and Intel, OPEA—one of the Linux Foundation’s Sandbox projects, an incubator program—counts among its members such heavyweights as Intel, IBM-owned Red Hat, Hugging Face , Domino Data Lab, MariaDB and VMWare.
What exactly could they build together? Haddad hints at a few features, such as “optimized” support for AI toolchains and compilers, allowing AI workloads to run on different hardware components, as well as “heterogeneous” pipelines for recovery production augmented (RAG).
RAG is becoming increasingly popular in business genetic AI applications, and it’s not hard to see why. The responses and actions of most generative AI models are limited to the data they have been trained on. But with RAG, a model’s knowledge base can be extended to information beyond the original training data. RAG models refer to this external information—which can take the form of proprietary corporate data, a public database, or some combination of the two—before generating a response or executing a task.
A diagram explaining RAG models.
Intel offered a few more details of its own Press release:
Businesses are challenged with a do-it-yourself approach. [to RAG] because there are no de facto standards in all components that allow enterprises to select and develop RAG solutions that are open and interoperable and that help them get to market quickly. OPEA intends to address these issues by working with industry to standardize components, including frameworks, architecture designs, and reference solutions.
Evaluation will also be a key part of what OPEA is dealing with.
On his GitHub warehouseOPEA proposes a rubric for grading production AI systems along four axes: performance, features, reliability, and “enterprise-grade” readiness. Implementation as defined by OPEA refers to “black box” benchmarks from real-world use cases. Characteristics is an assessment of a system’s interoperability, deployment options, and ease of use. Trustworthiness examines the ability of an AI model to guarantee “robustness” and quality. And operational readiness it focuses on the requirements for running a system without significant problems.
Rachel Roumeliotis, director of open source strategy at Intel, says that OPEA will work with the open source community to offer rubric-based testing, as well as provide assessments and grading of genetic AI deployments upon request.
OPEA’s other efforts are a bit up in the air right now. But Haddad used open model development capabilities along the lines of Meta’s expanding Llama family and Databricks’ DBRX. Toward this end, in the OPEA repository, Intel has already contributed reference implementations for an AI-enabled chatbot, document digest, and code generator optimized for Xeon 6 and Gaudi 2 hardware.
Now, OPEA members are clearly invested (and have interests, for that matter) in building tools for business AI. Cloudera recently started partnerships to build what it touts as an “AI ecosystem” in the cloud. Domino offers a application suite to build and control business artificial intelligence. And VMWare — geared toward the infrastructure side of enterprise AI — launched last August new “private artificial intelligence” computing products.
The question is whether these sellers will actually work together to build multiple compatible AI tools under OPEA.
There is an obvious benefit to this. Customers will happily leverage multiple vendors depending on their needs, resources and budgets. But history has shown that it is very easy to tend towards vendor lock-in. Hopefully that’s not the end result here.
