Seemingly identical AI large models, why do they ultimately head in completely different directions?

On the surface, whether it's Optimus, Neo, or Phoenix, they may all originate from the same model weights, the same instruction set, run the same LLM backend, and be constrained by the same conditions. But this is just the starting point.

The real difference comes later—different fine-tuning strategies, diverse application scenarios, and their respective ecosystem positioning. Just like open-source projects forked from the same base code, initially almost identical, but as different teams iterate and optimize, and market demands drive them, they gradually evolve their own unique features and advantages.

In an era of intense AI competition, how the same basic infrastructure can develop differentiated capabilities is a question every model innovator is pondering.