The big picture: In recent days, the AI community has witnessed the emergence of a new generation of AI models, heralding a significant leap in capabilities and potential applications. Claude 3.7 and Grok 3 are pushing the boundaries of what AI can achieve, particularly with complex tasks, mathematics, and coding. These Gen3 models represent a quantum leap in computing power utilized during training, according to a post written by Ethan Mollick in the Substack newsletter One Useful Thing. Grok 3, developed by Elon Musk's xAI, is the first known model to employ an order of magnitude greater computing power than its predecessor, GPT-4. Claude 3.7, for its part, showcases substantial performance improvements and introduces new coding and reasoning capabilities.The advancements in these models are underpinned by two critical "Scaling Laws" identified by OpenAI. The first law, illustrated on the left-hand side of the graph, demonstrates that larger models trained with more computing power exhibit enhanced capabilities. This relationship is not linear; typically, a tenfold increase in computing power is required to achieve a linear improvement in performance.Image credit: Ethan MollickThe scale of computing power involved in training these new models is staggering. Gen3 models utilize over 10^26 FLOPS during training, equivalent to running a modern smartphone for 634,000 years or the Apollo Guidance Computer for 79 trillion years.The second Scaling Law, represented on the right-hand side of the graph, reveals an intriguing phenomenon: AI performance can be improved by allowing the model more time to process information during problem-solving. // Related StoriesThis discovery has led to the development of "Reasoners," AI systems that can allocate additional computing resources to tackle complex problems more effectively according to Mollick.These advancements are not merely academic; they have profound implications for real-world applications. For instance, Claude 3.7 has demonstrated the ability to create interactive 3D visualizations of complex academic concepts and generate functional code through natural language conversations.In one example, the AI produced an interactive time machine artifact complete with pixel graphics, showcasing its capacity for creative and technical tasks.However, Mollick notes that while these systems are impressive, they are not infallible. They still make mistakes and have limitations. Nevertheless, the rapid pace of improvement suggests that AI capabilities will continue to expand.As they do, they challenge the prevailing "automation mindset" in corporate environments, which often view AI primarily as a tool for streamlining existing processes. Instead, according to Mollick, these new models invite a fundamental rethinking of what's possible, positioning AI as a potential intellectual partner capable of tackling complex analytical tasks, creative work, and even research-level problems.This shift will require a new approach to AI integration in organizations. Leaders must move beyond task automation to capability augmentation, asking not just what can be automated, but what new capabilities can be unlocked.As these models become more accessible, Mollick urges individuals and organizations to explore their capabilities firsthand. Both Claude 3.7 and Grok 3 offer unique features and strengths, with Claude 3.7 providing code execution capabilities and Grok 3 offering a broader set of capabilitiess, including deep research options.