The Battle for Open-Source AI in the Wake of Generative AI

Open-source AI is rapidly reshaping the software ecosystem by making AI models and tools accessible to organizations. This is leading to a number of benefits, including accelerated innovation, improved quality, and lower costs.

According to the 2023 OpenLogic report, 80% of organizations are using more open-source software compared to 77% last year to access the latest innovations, improve development velocity, reduce vendor lock-in, and minimize license costs.

The current landscape of open-source AI is still evolving. Tech giants such as Google (Meena, Bard, and PaLM), Microsoft (Turing NLG), and Amazon Web Services (Amazon Lex) have been more cautious in releasing their AI innovations. However, some organizations, such as Meta and other AI-based research companies, are actively open-sourcing their AI models.

Moreover, there is an intense debate over open-source AI that revolves around its potential to challenge big tech. This article aims to provide an in-depth analysis of the potential benefits of open-source AI and highlight the challenges ahead.

Pioneering Advancements – The Potential of Open-Source AI

Many practitioners consider the rise of open-source AI to be a positive development because it makes AI more transparent, flexible, accountable, affordable, and accessible. But tech giants like OpenAI and Google are very cautious while open-sourcing their models due to commercial, privacy, and safety concerns. By open-sourcing, they may lose their competitive advantage, or they would have to give away sensitive information regarding their data and model architecture, and malicious actors may use the models for harmful purposes.

However, the crown jewel of open-sourcing AI models is faster innovation. Several notable AI advancements have become accessible to the public through open-source collaboration. For instance, Meta made a groundbreaking move by open-sourcing their LLM model LLaMA.

As the research community gained access to LLaMA, it catalyzed further AI breakthroughs, leading to the development of derivative models like Alpaca and Vicuna. In July, Stability AI built two LLMs named Beluga 1 and Beluga 2 by leveraging LLaMA and LLaMA 2, respectively. They showcased better results on many language tasks like reasoning, domain-specific question-answering, and understanding language subtleties compared to state-of-the-art models at that time. Recently, Meta has introduced Code LLaMA–an open-source AI tool for coding that has outperformed state-of-the-art models on coding tasks – also built on top of LLaMA 2.

Researchers and practitioners are also enhancing the capabilities of LLaMA to compete with proprietary models. For instance, open-source models like Giraffe from Abacus AI and Llama-2-7B-32K-Instruct from Together AI are now capable of handling 32K long input context lengths – a feature that was only available in proprietary LLM like GPT-4. Additionally, industry initiatives, such as MosaicML's open-source MPT 7B and 30B models, are empowering researchers to train their generative AI models from scratch.

Overall, this collective effort has transformed the AI landscape, fostering collaboration and knowledge-sharing that continue to drive groundbreaking discoveries.

Benefits of Open-Source AI for Companies

Open-source AI offers numerous benefits, making it a compelling approach in artificial intelligence. Embracing transparency and community-driven collaboration, open-source AI has the potential to revolutionize the way we develop and deploy AI solutions.

Here are some benefits of open-source AI:

• Rapid Development: Open-source AI models allow developers to build upon existing frameworks and architectures, enabling rapid development and iteration of new models. With a solid foundation, developers can create novel applications without reinventing the wheel.
• Increased Transparency: Transparency is a key feature of open-source, providing a clear view of the underlying algorithms and data. This visibility reduces bias and promotes fairness, leading to a more equitable AI environment.
• Increased Collaboration: Open-source AI democratized AI development, which promotes collaboration, fostering a diverse community of contributors with varying expertise.

Navigating Challenges – The Risks of Open-Sourcing AI

While open-source offers numerous advantages, it is important to be aware of the potential risks it may entail. Here are some of the key concerns associated with open-source AI:

• Regulatory Challenges: The rise of open-source AI models has led to unbridled development with inherent risks that demand careful regulation. The sheer accessibility and democratization of AI raise concerns about its potential malicious use. According to a recent report by SiliconAngle, some open-source AI projects use generative AI and LLMs with poor security, putting organizations and consumers at risk.
• Quality Degradation: While open-source AI models bring transparency and community collaboration, they can suffer from quality degradation over time. Unlike closed-source models maintained by dedicated teams, the burden of upkeep often falls on the community. This often leads to potential neglect and outdated model versions. This degradation might hinder critical applications, endangering user trust and overall AI progress.
• AI Regulation Complexity: Open-sourcing AI models introduce a new level of complexity for AI regulators. There are a number of factors to consider, such as how to protect sensitive data, how to prevent models from being used for malicious purposes, and how to ensure that models are well-maintained. Hence, it is quite challenging for AI regulators to ensure that open-source models are used for good and not for harm.

The Evolving Nature of Open-Source AI Debate

“Open source drives innovation because it enables many more developers to build with new technology. It also improves safety and security because when software is open, more people can scrutinize it to identify and fix potential issues”, said Mark Zuckerberg when he announced the LLaMA 2 large language model in July this year.

On the other hand, major players like Microsoft-backed OpenAI and Google are keeping their AI systems closed. They are aiming to gain a competitive advantage and minimize the risk of AI misuse.

OpenAI’s co-founder and chief scientist, Ilya Sutskever, told The Verge, “These models are very potent and they’re becoming more and more potent. At some point, it will be quite easy, if one wanted, to cause a great deal of harm with those models. And as the capabilities get higher, it makes sense that you don’t want to disclose them.” So, there are potential risks related to open-source AI models that humans cannot ignore.

While AIs capable of causing human destruction may be decades away, open-source AI tools have already been misused. For example, the first LLaMA model was only released to advance AI research. But malicious agents used it to create chatbots that spread hateful content like racial slurs and stereotypes.

Maintaining a balance between open AI collaboration and responsible governance is crucial. It ensures that AI advancements remain beneficial to society while safeguarding against potential harm. The technology community must collaborate to establish guidelines and mechanisms that promote ethical AI development. More importantly, they must take measures to prevent misuse, enabling AI technologies to be a force for positive change.

Want to enhance your AI IQ? Navigate through Unite.ai‘s extensive catalog of insightful AI resources to amplify your knowledge.