GPT-4, Anthropic’s Claude 3 Opus, and Google’s Gemini Advanced1 are three prominent AI models in the market. While there is ongoing debate about their superiority, each model has distinct personalities and strengths. All three models possess multimodal capabilities, enabling them to process images and perform various real-world tasks. However, these models lack adequate documentation and instructions for optimal usage. The use of context windows and Retrieval Augmented Generation (RAG) has extended the capabilities of these models, enhancing their understanding of new data. Additionally, the emergence of AI agents like Devin signals a potential shift in integrating AI into organizations. Despite the existence of advanced AI models, further advancements, such as larger context windows and improved AI agents, are expected in the near future.
| Signal | Change | 10y horizon | Driving force |
|---|---|---|---|
| Emergence of multiple GPT-4 class models | Increase in AI model options and capabilities | Continued development of more advanced AI models | Competition and advancement in AI technology |
| GPT-4 class models emulate human conversation well | AI models become more human-like in conversation | AI models will seem increasingly human-like and realistic | Desire to create more realistic and engaging AI experiences |
| Ability of AI models to “see” images | AI models gain multimodal capabilities | AI models can interpret and analyze visual content | Expansion of AI applications to visual tasks |
| Lack of instructions/documentation for LLMs | Need for better understanding and guidance | Improved documentation and guidelines for LLM usage | Demand for user-friendly and accessible AI technologies |
| Similar functioning of advanced AI models | Interchangeability of AI models | Users can swap out AI models and achieve similar results | Standardization and compatibility in AI technology |
| Increasing context windows and RAG capabilities | Enhanced context utilization in LLMs | AI models with larger context windows and better RAG | Need for contextual understanding and data customization |
| Development of autonomous AI agents | Shift towards autonomous AI programs | Integration of autonomous AI agents in various domains | Desire to enhance AI capabilities through automation |
| Anticipated release of GPT-5 and Gemini 2.0 models in future | Advancements in AI model capabilities | Release of more advanced and efficient AI models | Continuous innovation and improvement in AI technology |
| Challenges and considerations in choosing AI models | Need for better decision-making tools and resources | Improved tools and resources for selecting AI models | Simplifying the AI model selection process |