This article serves as a follow-up to my previous Substack post on the same subject. While the earlier piece primarily explored the issue from a safety perspective, this time I aim to examine it through the lens of economic principles. It appears evident from an economic standpoint that the open-source AI model runs counter to fundamental economic principles within the framework of free market capitalism. There may be those who contend that open-source AI transcends the importance of the capitalist system, suggesting its abandonment for what they perceive as greater good. However, for the sake of argument, let's proceed under the assumption that free market capitalism is our agreed-upon foundation.
Time stands as humanity's most precious resource on Earth. The Austrian economist Ludwig von Mises eloquently stated, "Man is subject to the passing of time. He comes into existence, grows, becomes old, and passes away. His time is scarce. He must economize it as he economizes other scarce factors." This scarcity of time naturally leads to the limited availability of human labor and talent, culminating in the rare presence of exceptionally gifted individuals. Capital, essentially the product of human labor, accumulates over time, making it a scarce and valuable representation of time itself. This can be considered a fundamental principle of economics.
Currently, the market for large language models (LLMs) sees a competition between open-source and closed-source variants within the same economic framework. As of this article, OpenAI's GPT-4 stands as the market leader among all models. According to Simon Willison in his article, all contender models are closed-source, with no open-source models matching their performance. Since the release of GPT-4 on March 14, 2023, its competitors have improved, yet none has surpassed it. With rumors of GPT-5 on the horizon , the gap between GPT-5 and its rivals is expected to grow, further expanding the divide between closed and open-source LLMs. This development has led to widespread frustration, underpinned by fundamental economic principles.
In the realm of free market capitalism, companies vie for scarce resources, whether it's the industry's brightest minds, substantial computational resources, or ample capital. Those firms that combine top-tier talent, significant computational capacity for model training, and substantial funding to support their growth emerge as leaders, exemplified by the success of GPT-4. This technological leadership translates into market dominance, evidenced by 80% of Fortune 500 companies adopting GPT-4, and the achievement of $2 billion in annual recurring revenue (ARR) by December 2023, merely nine months post-launch. Such milestones have not only attracted more of the industry's finest talents to OpenAI but also increased venture capital investment to boost computing capabilities and solidify technological supremacy. This positive feedback loop is formidable, showing no signs of abating soon. The message is clearer than ever: the competitive dynamics of free market capitalism are functioning as intended, and the success of OpenAI stands as a testament to this, regardless of the aspirations for open-source AI's triumph.
In the domain of open-source LLMs, Meta, a forefront player, has recently announced a significant investment in GPUs, aiming to bolster its position in the market. Mark Zuckerberg, Meta's CEO, has pledged to keep their new model, Llama 3, open-source, emphasizing the importance and potential of the technology: "This technology is so important and the opportunities are so great that we should open source and make it as widely available as we responsibly can." Yet, there's speculation that, true to Zuckerberg's pragmatic approach, the model might eventually shift to a closed-source model if it surpasses market competitors or if misuse concerns arise, as previously discussed in another Substack post of mine.
Furthermore, in a surprising move spurred by Elon Musk's legal actions against OpenAI and shared via a series of tweets on X (formerly Twitter), x.ai has also decided to open source its Grok-1 model. This decision highlights a growing trend towards open-source LLMs, although it's crucial to manage expectations. Neither Llama 3 nor Grok-1 currently matches the performance of OpenAI’s GPT-4, which remains a leader in the field more than a year after its release. Despite Meta and X.ai operating within the competitive landscape of free market capitalism, bolstered by substantial corporate resources and capital investment, their choice to adopt an open-source model may still outspeak loudly that they are not winning the market.
The contributions of open-source LLMs to the development of artificial intelligence (AI) are significant and cannot be overlooked. Entities such as Meta and X.ai, along with a myriad of smaller open-source model providers, offer crucial resources that enable individuals, hobbyists, researchers, and companies to tailor AI tools for specific applications, thereby reducing costs and safeguarding privacy within their own environments. These endeavors are integral to the broader ecosystem, benefiting a wide array of stakeholders. The future will undoubtedly see the emergence of innovative tools built on open-source LLMs, continuing the legacy of contributions to the ecosystem similar to what Linux has achieved. Nonetheless, the notion that open-source LLMs might supplant closed-source models confronts the harsh reality of competition within the free market system. True innovation, driven by the economic principle that highlights the scarcity of time and human endeavor, invariably bears a cost.
Investment in AI reached new heights in 2023, yet in recent months, venture capitalists (VCs) have become more cautious about making large investments. The turbulence within AI startups is largely attributed to their performance metrics, financial health, and customer base in an increasingly competitive landscape. The departure of top executives from Inflection, as reported by Forbes, may signify the onset of a broader recalibration of expectations among VCs who have been generous with their funding over the past year. Similarly, the resignation of the CEO from Stability AI could reflect additional competitive pressures within the AI sector. Smaller open-source AI startups are particularly vulnerable in this intense competitive environment, many lacking a solid business model to generate sustainable revenue. While some open-source entities have devised strategies to monetize AI services, the viability of these revenue streams to meaningfully offset costs remains a significant concern.
As the competitive landscape persists, it's evident that numerous companies will be phased out. Those lacking advanced technology, unable to generate revenue, or without sufficient capital to support their operations will inevitably fall behind. A significant portion of these will be companies relying on open-source models, as their inherent nature places them at a competitive disadvantage. Only a select few open-source model companies will manage to endure and offer their distinct contributions to the AI industry. Meanwhile, closed-source models are poised to emerge as the predominant forces, drawing in the most skilled AI scientists, securing the greatest amount of capital, and producing the majority of revenue. Consequently, closed-source AI will drive technological progress in numerous facets of the AI field.
As an investor, my perspective extends beyond the immediate conclusions. Should open-source LLMs fail to make a significant impact in the long term, it becomes apparent that the broader ecosystem surrounding open-source innovation may not yield substantial value. Thus, it is crucial to exercise caution and potentially steer clear of investment opportunities in open-source models and development tools.
(This is an ongoing observation on an extremely important topic; Comments are always welcome; Will update as the LLMs landscape evolves; With the help of ChatGPT)