Opinions expressed by Entrepreneur contributors are their own.
Field of HE is progressing rapidly. Major companies continue to launch new basic models. However, there is no clear definition of a fully open AI model. Many models claim to be “open”, but only a subset of components are open source and use restrictive licensing for the rest. This creates a partial aperture spectrum. For example,
- one can publish the architecture and weights of a model, but not the data and training code.
- one may release the trained weights under a license that prohibits commercial use or restricts derivative work,
- or one can release the trained weights under a non-restrictive license but the code under a restrictive license.
This ambiguity about what is truly “open” gets in the way the progress of AI adoption, creating products and services for the end user. Creates legal risks for entrepreneurs who may inadvertently violate the terms of partially open models. We need a clear framework for assessing the open nature of the model. Such a framework should help AI entrepreneurs, researchers, and engineers make informed decisions about which models to use, build derivative works, and make contributions to them.
An example
Let's consider a hypothetical The beginning of AI called “end-onether-chat-bot.” They are developing an AI chatbot to improve customer support responses. They used a hypothetical pre-trained language model called “lam-stral” to speed up development. The authors of “llam-stral” have published a paper on arXiv describing the architecture and performance. They have made the weights available for download.
The engineers of yet-another-chat-bot use “lam-stral” in their prototype chatbot, but later discover that the license explicitly prohibits commercial use and the creation of derivative works. Also, the training data and the code used for training have not been published. They are now exposed to legal risks and potential IP infringement issues.
The right thing to do would have been for the “lam-stral” to stick to it Pattern Opening Frame and use a standard open license such as Apache 2.0 for the code and CC-BY-4.0 for the weights and datasets. It would have been pretty obvious for the end-other-chat-bot startup to use it commercially and build on it.
There is a need for a framework that defines the integrity and openness of models for effective reproducibility, transparency and usability in AI. Using something like Pattern-opening frame published by GenAICommons would be useful for both model creators and consumers to understand which are the key artifacts, which are open and which are not. A fully open model would release all components, including training data, code, weights, architecture, technical report, and evaluation code, all under permissive licenses.
Connected: Scarlett Johansson asks why ChatGPT sounds like her
Components of an AI model
By releasing all artifacts and components related to a large language model under permissive licenses, creators can claim that their models are truly and fully open. This promotes transparency, reproducibility and collaboration in the development and application of large language models
Some of the essential components are as follows:
- Training data: Data used to train the large language model.
- Data preprocessing code: Code used to clean, transform and prepare the training data.
- Model architecture: AI model design and structure, including its layers, connections, and hyperparameters.
- Model parameters: The learned weights and biases of the trained AI model.
- Training code: The code used to train the AI model, including the training loop, optimization algorithm, and loss functions.
- Evaluation code: Code used to evaluate the performance of the trained AI model on the validation and test datasets.
- Evaluation data: Data used to evaluate the performance of the trained AI model.
- Model of documentation and technical report: Detailed documentation of the AI model, including the purpose, architecture, training process, and performance metrics. Academic paper or technical report describing the AI model, its methodology, results and contributions to the field.
The more artifacts that are open and licensed with permission, the more open the model.
Connected: OpenAI and Meta models will soon have 'reasoning' capabilities
Truly open models accelerate innovation
Access to reality open AI models it levels the playing field for AI entrepreneurs and helps unleash innovation. They would use the latest models and datasets instead of building every component from scratch. This would help them prototype ideas faster and prove performance, speeding up time to market.
Instead of spending time and resources to reinvent the wheel and recreate core capabilities, AI entrepreneurs can now focus on domain-specific challenges and identify ways to add value. Open licenses used by models in accordance with Model Opening Framework (MOF) also provide confidence that entrepreneurs can legally use designs in commercial products and services.
There will be no worries about the risk of IP infringement claims or unexpected changes to licensing terms. Access to all training data and code under non-restrictive licenses helps entrepreneurs control the origin of the model, ensuring compliance with regulations.
Additionally, an engineer can examine data sets for possible biases. Developers will be able to find performance bottlenecks and improve performance as they will have access to the entire codebase. This can help transfer the model to different environments and improve maintainability over time. Thus, fully open models reduce the barriers to building AI-powered products and services and move the innovation needle.