AI Investment Trends: Key Themes and Opportunities in 2024

The Evolving AI Landscape: Key Themes of 2024

At NeurIPS 2024, Sarah Guo and Pranav Reddy from Conviction Capital shared their insights on the current state of AI startups. They highlighted five crucial themes that defined the AI landscape in 2024, and explored their implications for future investments. These themes offer a glimpse into the rapidly changing dynamics of artificial intelligence.

Intensified Competition in Foundation Models

The year 2024 witnessed unprecedented competition among foundation models. This intense rivalry is pushing the boundaries of what's possible, with new models constantly emerging and challenging the status quo.

The Rise of Open-Source Models

Open-source models have rapidly gained competitiveness, even surpassing closed-source models in certain aspects. This trend is empowering developers and researchers with greater access to powerful AI tools. The rise of open-source models is democratizing AI technology.

Cost-Effective Small Models

Small models have made remarkable strides in performance while maintaining lower costs, leading to a more favorable price-to-performance ratio. The ability to deploy smaller, more efficient models is making AI more accessible and practical for a wider range of applications.

Multimodal Technology Breakthroughs

Multimodal technology is emerging as a significant area of development, offering users entirely new interactive experiences. Combining various data types such as text, images, and audio is leading to more versatile and intuitive AI systems.

New Paradigms in Scaling Law

Despite the limitations of Scaling Law, new expansion paradigms are emerging, offering fresh perspectives on the future of AI development. These breakthroughs are pushing the boundaries of AI capabilities and opening up new avenues for research and development.

The Competitive Dynamics of Base Models

The competitive landscape for base models has transformed significantly in 2024. Data from Chatbot Arena reveals that while OpenAI's models were once dominant, other companies like Google have introduced competitive alternatives. The progress in open-source models is also providing businesses with more choices when selecting APIs.

Declining Token Consumption for OpenAI

OpenAI's token consumption, which accounted for nearly 90% of the total at the end of November last year, has fallen to almost 60% within a year. This indicates that users are exploring different models and are not solely reliant on OpenAI's offerings. The shift in token consumption reflects a more diversified AI ecosystem.

The Ascendancy of Open-Source Models

Independent evaluations by the SEAL rankings demonstrate that open-source models excel in areas such as mathematical ability, instruction following, and adversarial robustness. Notably, the Llama model has secured a position among the top three. The growing performance of open-source models is challenging the dominance of proprietary models.

The Rapid Advancement of Open-Source Models

Open-source models have exhibited remarkable progress across numerous dimensions. MMLU evaluations indicate that some open-source models with smaller parameter sizes are approaching the performance of the most advanced models. For example, the Mistral-7b model, which scored around 60 in MMLU a year ago, has been surpassed by the Llama 8B model, which has improved its score by more than 10 points. This level of improvement is indicative of the rapid development of open-source AI.

The Cost-Effectiveness of Smaller Models

The performance gap between small and large models is narrowing, and the cost of AI is decreasing substantially. The API costs of OpenAI's flagship models have decreased by approximately 80-85% over the past year to a year and a half. This reduction in cost is making AI more accessible to startups and smaller enterprises, which were previously priced out of using advanced models.

Reduced Costs for AI Application Development

The reduced costs of AI mean that building applications similar to Notion or Coda now requires only a few thousand dollars in token costs. This makes AI application development more financially viable for a broader range of businesses and developers. The decrease in costs is a significant catalyst for AI innovation.

Multimodal Technology: The Future of AI

Multimodal technology is becoming a critical development direction in the AI field. The emergence of new modalities such as low-latency voice, execution capabilities, and video is creating entirely new interactive experiences. This is pushing the boundaries of how users interact with AI systems.

Enhanced Voice Interaction Experiences

Low-latency voice is not just a feature; it represents an entirely new interaction experience. The ability to have seamless, real-time conversations with AI systems is significantly improving usability and user satisfaction. Voice interaction is becoming a more natural and intuitive way to interact with technology.

The Growing Execution Capabilities of AI

The Computer Use capabilities of Claude and the code execution features introduced by OpenAI in Canvas demonstrate the enhanced execution capabilities of AI. These advancements are providing users with more powerful and versatile AI tools. The increasing ability of AI to execute tasks is expanding its applications.

Scaling Law Breakthroughs

While Scaling Law does have its limitations, new expansion paradigms are emerging. OpenAI appears to have found ways to overcome these limitations, using techniques such as RL self-play to enhance model performance. These breakthroughs are crucial for the continued development and advancement of AI.

A More Rational AI Investment Environment

Despite some claims that the AI field is experiencing a bubble, the reality is that the funding flowing into foundation model labs is relatively high, while the funding directed to actual operating companies is more reasonable. This suggests a shift towards a more balanced investment approach.

The Underestimated Value of the Application Layer

In the past, it was commonly believed that the foundation models were more valuable, and the application layer was not. However, the opportunities within the AI ecosystem are vast, and the application layer has considerable potential. This shift in perspective is creating new investment opportunities in the AI space.

Significant Opportunities for Startups

The opportunities in the AI ecosystem extend beyond foundation models. The application layer offers enormous potential, with numerous innovations emerging, different types of models excelling in various areas, fierce market competition, and flourishing open-source projects. This environment is ripe for new startups to innovate and disrupt the market.

Service Automation

AI can automate many repetitive tasks, reducing costs and improving efficiency. This automation is transforming various industries by streamlining operations and freeing up human resources for more complex tasks. The use of AI in service automation is increasing productivity and efficiency.

Improved Search Paradigms

AI can help users search for and obtain information more effectively, providing more personalized experiences. AI-powered search engines are more intuitive and can provide more relevant results than traditional search methods. The ability of AI to understand context and user intent is transforming the search experience.

Democratization of Skills

AI is democratizing various skills, enabling more people to participate in creation and innovation. This is empowering individuals with the tools they need to develop new skills and contribute to various fields. The democratization of skills through AI is fostering innovation and creativity.

Investment Focus in the AI Wave

AI Infra (computing power and data) is a significant investment direction in the AI wave. As AI technology advances, the demand for expert data and more types of data is continuously increasing. This growing demand is creating investment opportunities in data infrastructure and related technologies.

The Dawn of "Software 3.0"

In summary, we use the term "Software 3.0" to describe these transformative changes. We believe this is a full-stack rethink that will provide significant advantages to the next generation of companies. The pace of change favors startups, as they can be more agile and adaptable to new technologies.

Rethinking Products and Infrastructure

We need to rethink how we design products and build infrastructure to accommodate the new AI paradigm. This requires a fundamental shift in how we approach software development and technological innovation. The changes brought about by AI demand a new perspective on product design and infrastructure.

Immense Technological and Economic Opportunities

AI presents immense technological and economic opportunities that are worth seizing. The potential for AI to transform industries and create new markets is vast, and investors are eager to capitalize on these opportunities. The economic impact of AI is expected to be substantial.

The Battle Between Startups and Giants

Will the fruits of victory ultimately flow to startups or existing giants? While giants have distribution channels and data advantages, startups can compete with better products and innovative business models. The ability of startups to be more nimble and innovative provides them with a competitive edge.

The Innovator's Dilemma

Existing companies may be limited by the innovator's dilemma, while startups can challenge existing models with new user experience paradigms and code generation. Startups are well-positioned to disrupt established markets by leveraging new technologies and approaches.

The Importance of Data

Startups need to think about what kind of data they need to improve product quality, rather than simply relying on existing company data. The ability to acquire and utilize relevant data is crucial for the success of AI startups. The strategic use of data is a key differentiator in the competitive AI landscape.