21 Sep 2023
by Preeti

How AI is Increasing its capabilities with open-source foundation models

Guest blog by Preeti, Director of Consulting at Ve3 Global. Part of techUK's #SuperchargingUKTech week 2023.

Thursday 15.png 1

There’s no doubt that AI models such as ChatGPT have made our lives simpler. However, these have a substantial limitation: due to a lack of breadth in the datasets upon which they have been trained, the resultant model is only able to target a specific task. The implications are simple: narrow datasets mean narrow applications. 

This is far from ideal. In a perfect world, we would want to see AI models designed in such a way that one model, after being trained on a broad dataset, replaces many task-specific models. Such a model would speed up the development and adoption of AI as task-specific AI models demand greater amounts of time and money. It is simply not feasible to dedicate this amount of resources when running multiple task-specific businesses. 

Switching to foundation models  

Foundation models are large, pre-trained artificial intelligence models that have been trained on vast amounts of text data from the internet. The use of such models could save organisations significant amounts of time and money, driving faster rates of AI development and adoption in the process. In other words, the use of foundation models enables AI to scale, and fast. 

Introducing Open-Source Foundation Models 

An open-source foundational model serves as a foundation for training AI models with general data, allowing adaptability to specific requirements. These models enable AI systems to perform intermediate computations, leading to data standardisation and the attainment of desired outcomes. 

For instance, consider a neural network built upon a foundational model that analyses millions of image datasets to recognise a specific query, such as a cat. By discerning patterns in pixel arrangements within cat images, the system can effortlessly learn from these experiences. Consequently, the system can autonomously acquire and improve its skills over time using a diverse range of datasets, without being reliant on task-specific labeled datasets, which can be both time-consuming and costly. 

How Open-Source Foundation Models are Enhancing AI Capabilities? 

Outputs are Closer to Expectations 

When selecting an open-source AI foundation model, it can yield more balanced and realistic results. Consider a startup working on an object prototype system. Initially, they used DALL-E, a closed-source foundation model, which generated artistic images. Upon switching to Mid-Journey, another closed-source foundation model, the images appeared animated. 

However, when they employed Stable Diffusion, an open-source foundation model, the images looked much more realistic and natural. The choice of model depended on the intended purpose. 

The presence of open-source foundation models has facilitated quick AI adoption for enterprises, sparing them the effort of building infrastructure from scratch. They can swiftly achieve desired outcomes at a fraction of the cost, thanks to a vast, open-source data repository. However, when these open-source foundation models become unavailable. Would deploying AI in your business be as convenient as it is today? 

Optimum Training Environment 

Why are 95% of global leaders favouring open-source foundational models when it comes to establishing the right infrastructure? The answer lies in the significant advantages they offer. 

Creating custom models from scratch can be an arduous task. Despite the availability of various generic APIs, they often fall short of delivering the specific use-case effects necessary to achieve desired results. In contrast, adopting open-source foundation models can dramatically enhance AI capabilities. It involves downloading readily available, pre-trained models and tailoring them to suit specific use-cases, a far more streamlined process. 

For instance, ClearML has embraced a wide array of open-source software (OSS) to enhance user interfaces, backend functionality, and various other components. They have also introduced open-source continuous integration and continuous delivery (CI/CD) workflows following the same model. These open-source solutions have accelerated AI development by fostering real-time community collaboration and facilitating timely adjustments. 

How Open-Source Foundation Models Help AI Overcome Challenges? 

Envision an organisation utilising AI-recruitment software reliant on learning datasets from a specific geographical area steeped in racial biases. The repercussions of such an AI implementation are profound, manifesting as bias, ethical concerns, and irrational decision-making. This underscores the limitations of legacy AI data models, with far-reaching implications: 

  1. Geopolitical 

  1. Technological 

  1. Social 

  1. Environmental 

  1. Psychological 

These issues underscore the pressing need for a shift towards open-source foundation models, which offer a remedy. By embracing a decentralised open-source AI model, existing AI capabilities can be elevated in several ways: 

  1. Economic Solutions: AI can integrate with Decentralised Ledger Technology (DLT) to create a platform employing smart contracts. These contracts enhance automation and facilitate the development of next-generation decentralised open-source AI solutions, fostering trust and driving adoption. 

  1. Technological Solution: As AI adoption becomes increasingly pervasive and data utilisation skyrockets, a decentralised AI platform built upon an open-source foundation model emerges as a preferred choice. It bolsters the security of critical IT infrastructure and data, guarding against cyber threats. 

Looking ahead, it's crucial to acknowledge that mainstream AI adoption entails both the responsible use and potential misuse of data. Therefore, solutions must be devised to harness datasets effectively, not only for training diverse AI models but also for ensuring their safety and reliability. 

Key Tools for Building the Next Gen AI Models 

Open-source foundational training models play a pivotal role in cultivating sustainable AI models, primarily by mitigating training biases. Consider the case of Amazon's AI training models, which may exclude data portraying a negative brand narrative. In contrast, employing open-source training models like TensorFlow, Model Zoo, or Hugging Face removes biases and broadens the scope for AI models to train on a diverse range of data. 

Why is this significant? 

Because incorporating all experiences, both positive and negative, encountered by brands, consumers, and stakeholders enriches the training process, enabling AI tools to enhance their capabilities significantly. 

IBM has applied for patenting of its 9,130 AI models which shall be used by more than 44% of the global companies in some way or the other in their operational process. With such a huge intake of companies relying on AI to improve their business process, it would be harrowing to see AI software’s getting trained by a single source which can be used as a brilliant tool for setting agendas and spreading false propaganda in a subtle way. 

Conclusion 

The rapid advancements in AI, driven by open-source foundation models, have transformed various industries. These models provide a strong foundation for developers and researchers to create specialised AI solutions. As AI continues to progress, the open-source community plays a vital role in fostering collaboration and innovation, expanding AI's possibilities. VE3 specialises in leveraging open-source foundation models to help organisations harness AI’s power. Our experienced team designs and deploys customised AI models tailored to your unique needs, ensuring seamless integration with your existing infrastructure. Partnering with VE3 keeps your organisation at the forefront of AI, streamlining operations and enabling data-driven decision-making.  


techUK – Unleashing UK Tech and Innovation 

innovation_icon_badge_final.png

The UK is home to emerging technologies that have the power to revolutionise entire industries. From quantum to semiconductors; from gaming to the New Space Economy, they all have the unique opportunity to help prepare for what comes next.

techUK members lead the development of these technologies. Together we are working with Government and other stakeholders to address tech innovation priorities and build an innovation ecosystem that will benefit people, society, economy and the planet - and unleash the UK as a global leader in tech and innovation.

For more information, or to get in touch, please visit our Innovation Hub and click ‘contact us’. 

Upcoming Innovation events:

Latest news and insights:

Get our tech and innovation insights straight to your inbox

Sign-up to get the latest updates and opportunities from our Technology and Innovation and AI programmes.

 

Learn more about our Unleashing Innovation campaign:

Unleashing the Potential of UK Tech and Innovation.jpg

 



techUK – Unleashing UK Tech and Innovation 

innovation_icon_badge_final.png

The UK is home to emerging technologies that have the power to revolutionise entire industries. From quantum to semiconductors; from gaming to the New Space Economy, they all have the unique opportunity to help prepare for what comes next.

techUK members lead the development of these technologies. Together we are working with Government and other stakeholders to address tech innovation priorities and build an innovation ecosystem that will benefit people, society, economy and the planet - and unleash the UK as a global leader in tech and innovation.

For more information, or to get in touch, please visit our Innovation Hub and click ‘contact us’. 

Upcoming Innovation events:

Latest news and insights:

Get our tech and innovation insights straight to your inbox

Sign-up to get the latest updates and opportunities from our Technology and Innovation and AI programmes.

 

Learn more about our Unleashing Innovation campaign:

Unleashing the Potential of UK Tech and Innovation.jpg

 


 

 

Authors

Preeti

Preeti

Management Consultant, Ve3 Global