I have been experimenting with the first of a new generation AI models, Claude 3.7 and Grok 3, for the last few days. Grok 3 is the first model that we know trained with an order of magnitude more computing power of GPT-4, and Claude includes new coding and reasoning capabilities, so they are not just interesting in their own right but also tell us something important about where AI is going. Before we get there, a quick review: this new generation of AIs is smarter and the jump in capabilities is striking, particularly in how these models handle complex tasks, math and code. These models often give me the same feeling I had when using ChatGPT-4 for the first time, where I am equally impressed and a little unnerved by what it can do. Take Claude’s native coding ability, I can now get working programs through natural conversation or documents, no programming skill needed. For example, giving Claude a proposal for a new AI educational tool and engaging in conversation where it was asked to “display the proposed system architecture in 3D, make it interactive,” resulted in this interactive visualization of the core design in our paper, with no errors. You can try it yourself here, and edit or change it by asking the AI.

Full analysis : Claude 3.7 and Grok-3 are the first “Gen3” models with big gains in handling complex tasks, using 10x more compute than GPT-4-class models, and better reasoning.