后软件时代，胜出只有这两条路可走

Core Insights - The article emphasizes the need for founders to innovate beyond traditional AI applications and avoid merely adapting existing workflows to AI, which could lead to obsolescence in the face of advancing foundational models [2][12][20] Investment Landscape - The rise of generative AI marks the end of a homogenized software era, where companies often optimized for funding milestones rather than sustainable growth [2][3] - Founders are encouraged to learn from Richard Sutton's "bitter lesson," which highlights that simpler systems with more data and computational power can outperform specialized expertise [3][20] Model Economy - The "model economy" is emerging, where businesses focus on providing resources necessary for model evolution, such as computational power and training data [7][8] - Companies like Oracle and AI cloud firms are adapting by supplying computational resources to AI labs, with Oracle's contracts reaching nearly $500 billion [7][8] Areas of Growth - Significant opportunities are identified in four key areas: - Commoditization of computational power: Companies are developing models to match supply and demand for computational resources [8] - On-device AI: AI applications that run locally to address latency, privacy, and cost issues are gaining traction [9] - Data trading platforms: New companies are emerging to facilitate the exchange of specialized data needed for AI model training [10] - Security: A shift towards proactive security measures for AI systems is anticipated, focusing on identifying vulnerabilities before they are exploited [11] Post-Materialization Applications - Not all AI applications are doomed; those that leverage unique model characteristics to create new workflows are likely to succeed [12][13] - Successful applications will avoid merely "skin-deep" AI integration and instead explore what is possible with AI's unique capabilities [12][13] Future Opportunities - Three major opportunities for post-materialization applications include: - Coordinating multiple agents: Tools that utilize multiple models to enhance decision-making and reduce biases [14][16] - Large-scale simulations: AI's ability to run thousands of experiments in parallel is transforming research methodologies [17] - Continuous feedback loops: AI systems that learn from every interaction can optimize and predict user needs without human intervention [18][19]