Narendran also proposes alternatives to transformer models, such as state space models, which compress information and make resource usage appear more linear. He also recommends exploring small language models (SLMs) and caching results from previous LLM invocations. He concludes by stating that the future of AI will likely involve significant changes, but the period of frenzy is over, and executives are becoming more thoughtful and realistic about their AI strategies.
Key takeaways:
- The costs—financial and environmental—of Generative artificial intelligence (GenAI) and large language models (LLMs) are becoming a concern due to their resource-intensive nature.
- The fundamental drawback of AI is the hidden cost of scaling projects in terms of resources, requirements, and sustainability, especially with transformer models.
- Alternatives to transformer models, such as state space models, are being explored to make AI more tractable in terms of resource usage and to provide a more financially and environmentally sustainable solution.
- Business leaders should consider exploring the possibilities of small language models (SLMs) and caching results from previous invocations of LLMs to manage costs and improve efficiency.