Open-Source LLMs: The Growing Impact on Generative AI in the Enterprise

Open-source large language models (LLMs) are becoming increasingly powerful tools in the field of generative AI in the enterprise. While closed models still dominate the market, experts believe that open-source models have the potential to catch up and even surpass their closed counterparts. However, there is still a lack of public examples of open-source model deployments by established companies.

The Current State of Open-Source LLM Deployments

There are a limited number of established companies that have publicly announced their deployment of open-source LLMs in real business applications. These companies include Meta, Mistral AI, IBM, Hugging Face, Dell, Databricks, AWS, and Microsoft. From interviews with these companies, it is clear that there are several initial cases of open-source LLM deployments, but the numbers are still relatively low. Industry observers predict that the adoption of open-source models will pick up later this year.

One reason for the slow adoption of open-source models is that they were released later than closed models. Meta released the first major open-source model, Llama, in February 2023, while Mistral AI released Mixtral, the top-performing open-source LLM, in December 2023. As a result, examples of deployment are just now starting to emerge. However, open-source advocates believe that open-source models will eventually catch up and surpass closed models in terms of deployment.

The Limitations and Advantages of Open-Source Models

While there are some limitations to the current open-source models, such as the difficulty of contributing to model development, open-source developers have been able to create thousands of derivative models that are achieving parity with or even outperforming closed models in certain metrics. The availability of open-source models allows enterprise companies to have more control over their data and fine-tune models for specialized purposes.

On the other hand, closed models may have little value for private companies, as they often lack easy access to their own data. Companies are increasingly turning to open-source models and exploring open-source-based customer support and code generation applications to interact with their own custom code. The use of open-source models can save companies money, especially if they have access to their own infrastructure.

The Challenges of Open-Source LLM Deployments

While open-source models offer numerous advantages, there are challenges in their deployment. Enterprise companies need to consider data privacy, customer experience, and ethical implications before moving forward with LLM applications. Companies typically start with internal use cases and deploy proof-of-concepts before moving on to external use cases. The ability to switch between different open and closed models is crucial to mitigate risks and ensure flexibility.

Moreover, the dichotomy between open and closed models is increasingly blurred, as most companies use a combination of both types. Open-source models provide companies with more control over their data, while closed models offer convenience and reliability. Companies often choose open source when they need to control access to their data or fine-tune models for specialized purposes. In the long term, open source is likely to be more cost-effective due to the absence of additional IP and development costs.

Examples of Open-Source LLM Deployments

Several companies have successfully deployed open-source LLMs in their applications:

  • VMWare deployed the HuggingFace StarCoder model to improve developer efficiency.
  • Brave, a privacy-focused web browser startup, uses the Mixtral 8x7B model for its conversational assistant, Leo.
  • The children-friendly mobile phone company relies on Hugging Face’s open-source models to add a security layer to screen messages.
  • Wells Fargo uses open-source LLMs, including Meta’s Llama 2 model, for internal purposes.
  • IBM leverages open-source LLMs from Hugging Face and Meta for various applications, including HR, consulting, marketing, and AI-generated insights.
  • Perplexity, an AI-powered search engine start-up, uses open-source LLMs for response generation.
  • Redpajama, Deci, Microsoft, AWS, and Google are among the companies that have developed open-source LLMs for enterprise usage.

While these examples provide a glimpse into the impact of open-source LLMs in the enterprise, it is important to note that many companies choose to keep their usage of open-source LLMs private. The expanding landscape of open-source LLMs makes it challenging to track the full extent of enterprise usage.

Leave a Reply

Your email address will not be published. Required fields are marked *

Related Posts