AXA Report – US Market Watch on Large Language Models and its impact on AXA by Raphael Ancellin

Key Takeaways Summary:

1. Large Language Models (LLMs) are likened to a new tech platform, comparable to the impact of the iPhone, showcasing unprecedented reasoning capabilities and competition with humans in specific tasks.
2. Despite a 26.7% reduction in AI private investment since 2021, Generative AI firms like OpenAI, Cohere, and Anthropic secure substantial funding and form partnerships with cloud providers. Big tech companies (Google, Meta) are also developing their own models.
3. A KPMG survey indicates that 60% of executives foresee barriers like talent acquisition, cost, and data privacy delaying the implementation of their first generative AI solution by 1-2 years. Tech investors predict major financial firms adopting LLM-powered services.
4. Cloud-based AI infrastructure is favored due to lower initial costs, but some companies consider on-premises options for extensive inference or training costs.
5. Closed-source models delivered via API as "AI as a Service" maintain superior performance, but open-source alternatives are progressing rapidly with sufficient accuracy for common use cases.
6. Bloomberg is a rare example of a company training its own large model. Others prototype using existing models, categorizing use cases into core AI, Industry AI, and commodity.
7. Regulatory compliance and data privacy concerns lead companies like Samsung, Verizon, and major US banks to impose restrictions on ChatGPT use.
8. The insurance industry must address new LLM-related risks, such as fraud and damages by autonomous AI agents, through specific exclusions or price adjustments.
9. Similar to the steam engine's impact, focus shifts from the technology (the model) to defining applications and business models. A product management approach prioritizing use cases and user pain points over technology becomes crucial