A Most Peculiar Speculation: Kite and the AI Bubble

This Kite, it seems, has taken flight with remarkable alacrity. Launched but a season ago, it has already ascended a staggering 205% in the current year (as of the tenth of March), placing it amongst the hundred most substantial of its digital brethren, with a market capitalization of some $513 million. One cannot help but observe the crowd gathering, eager to catch a glimpse of this ascending star. But is it genuine brilliance, or merely a reflection of the prevailing frenzy?



![A neural network reinforcement learning from human feedback (RLHF) model, comprising approximately 4,800 parameters, demonstrates the fastest convergence toward optimal performance, while a structurally simpler model-defined by only four parameters-exhibits high initial variance but ultimately achieves complete recovery at [latex] K=5,000 [/latex].](https://arxiv.org/html/2603.08956v1/ch08_rlhf/sims/gridworld_sample_complexity.png)



