The Future of Generative AI: Intense Competition and Slowing Performance Gains
Advances in large language models are leading to intense competition in the generative AI space, with concerns over commoditization and diminishing returns on performance gains.
Generative artificial intelligence (AI) has captivated the corporate world for over a year, with significant advancements in the field. However, the rapid progress in large language models (LLMs), which can produce and summarize text, may be starting to slow down. The technology developed by industry leaders like OpenAI, Google, Cohere, and Anthropic may not be as unique as initially thought, raising concerns about increased competition. This article explores the current state of generative AI, the challenges faced by companies in the space, and the potential implications for the future.
The Rise of Large Language Models
San Francisco-based OpenAI gained significant attention in November 2022 when it released ChatGPT, powered by a large language model. Previous iterations of LLMs produced incoherent and rambling text, but today’s models have shown impressive fluency. Google’s release of its Gemini suite of LLMs in December showcased the challenges of further progress. While Google’s Gemini model outperformed OpenAI’s GPT-4 in some benchmarks, the margin of difference was relatively small. This suggests that LLMs may become commoditized, with little differentiation between competitors.
Challenges and Legal Problems
Despite the advancements in LLMs, challenges remain. One significant issue is the propensity for LLMs to hallucinate or fabricate information. Additionally, generative AI companies face legal problems related to training on copyrighted material. Striking licensing deals with content providers is one solution, but it could impact profit margins. Gary Marcus, an emeritus professor of psychology and neural science, believes that companies in this space may be overvalued, and a recalibration could occur in the coming years.
Diminishing Returns on Scale
The progress in LLMs has largely been attributed to scale, involving vast amounts of training data and computing power to build complex models with billions of parameters. However, experts suggest that there are diminishing returns on increasing the number of nodes in LLMs. Alok Ajmera, CEO of Prophix Software Inc., notes that adding more data and computing power does not necessarily lead to more interesting outputs. While progress is not ending, the rate of efficiency and performance gains in LLMs is slowing down.
Increasing Competition and Open-Source Models
The number of LLMs available for corporate use is continually growing. In addition to proprietary developers like OpenAI, there is a thriving ecosystem of open-source LLMs that can be used for commercial purposes. New entrants, such as Mistral AI and Meta Platforms Inc., are also entering the space. Meta’s open-source push aims to prevent proprietary technology from dominating the market and commoditize generative AI. This increasing competition and availability of open-source models give companies the flexibility to switch providers easily.
Competing on Different Attributes
As LLM performance levels out, developers will need to compete on different attributes to attract customers. Toronto-based Cohere, for example, emphasizes the privacy and security benefits of its technology, addressing concerns of business users. Cost is also emerging as a crucial factor, with open-source models having an advantage in this regard. Many Canadian companies are opting for open-source models to leverage cost savings and pass them on to customers. However, a hybrid approach that combines different technologies is also expected to be prevalent.
Conclusion:
Generative AI has witnessed significant advancements in the form of large language models. However, the performance gains may be starting to slow down, leading to increased competition among companies in the space. The commoditization of LLMs and challenges related to hallucination and legal issues pose further obstacles. As the market evolves, companies will need to differentiate themselves based on attributes such as privacy, security, and cost. The availability of open-source models provides flexibility, but a hybrid approach may be the most effective strategy. As the generative AI landscape continues to evolve, organizations are advised to work with multiple players to navigate this rapidly changing field.