Sign up for our daily and weekly newsletters to get the latest updates and exclusive content on industry-leading AI developments.
Microsoft surprised everyone this morning by hiring Sam Altman and Greg Brockman, the former CEO and President of OpenAI, respectively. This strategic decision seems like Microsoft’s attempt to recover from the chaos that hit OpenAI just before the weekend when its board of directors decided to fire Altman.
However, the ultimate outcome of the OpenAI upheaval is still unknown. Many researchers have already left, and hundreds of employees, including top executives, are rebelling against the board’s decision. The future relationship between Microsoft and OpenAI is also uncertain since Microsoft plans to start an internal research division with Altman and Brockman, which will likely compete with OpenAI.
One thing is clear: OpenAI will never be the same, nor will its products like ChatGPT and its API platform. This disruption highlights the ever-changing nature of the cutting-edge AI industry. Scientists, engineers, and philosophers will continue debating the risks of advanced AI systems and the existential threats posed by artificial general intelligence (AGI).
Such conflicts will probably happen again, especially in AI labs trying to balance research and product development. Because of this, companies that have built products and applications based on OpenAI’s platform will need to reevaluate their strategies as the company’s future remains uncertain.
In this scenario, the market for open-source models might emerge as the biggest winner. Unlike closed-source systems like OpenAI’s platform, open-source models provide users full control and responsibility. They don’t have single points of failure, like an API server or a conflicting board that can’t decide whether to speed up product launches or slow down to assess risks.
Over the weekend, more than 100 OpenAI customers reportedly reached out to competitors like Anthropic, Google Cloud, Cohere, and Microsoft Azure. Enterprises can choose where and how to run open-source models, whether on their own servers, in a public cloud, or on a model-serving platform. Most major cloud providers offer ready-made access to open-source models like Llama 2, Mistral, Falcon, and MPT. This includes services from Microsoft Azure AI Studio and Amazon Bedrock, as well as numerous startups that offer easy access to hosted versions of open-source models. This wide range of options allows businesses to run models according to their existing infrastructure.
Furthermore, open-source models generally offer more stable performance compared to private models. Over the past year, there have been multiple reports of OpenAI model performance degrading (or changing) as the company continuously retrains, tweaks, and adjusts safeguard measures. These models are essentially black boxes within black boxes, making it hard to get consistent outputs.
In contrast, open-source models provide stable performance, with enterprises deciding when updates occur and what the safeguards are, avoiding panic-induced lockdowns from unexpected issues. The open-source model landscape is also advancing quickly, thanks to the collaborative efforts of researchers and developers.
There are numerous tools and techniques available now to customize open-source large language models (LLMs) for specific applications, which aren’t available for private models. Businesses can use methods such as quantization to reduce the costs of running models or low-rank adaptation to fine-tune them at significantly lower expenses, allowing thousands of models to operate on a single GPU. Open-source models can be adapted for various applications and budgets.
The problem with companies like OpenAI is their attempt to achieve two conflicting goals simultaneously: reaching AGI and producing profitable products to fund their research. These goals can sometimes be directly opposed, as demonstrated by the OpenAI saga.
In reality, most businesses don’t need AGI. They usually don’t require state-of-the-art models with trillions of parameters. What they need is a reliable foundation on which they can build stable LLM applications, even if it’s an LLM with only a few billion parameters. This is the opportunity that the open-source ecosystem offers. As the situation with OpenAI continues to develop, more companies are likely to turn to open-source LLMs.
Platforms like ChatGPT will still be useful for quick prototyping and exploring the possibilities of cutting-edge AI technology. But once the right application is found, businesses will be better off investing in technology that will remain stable regardless of the internal politics of the company that develops it.