A comprehensive report on extending backprop-free “NoProp” diffusion training from image classification to text generation, covering state-of-the-art diffusion LMs, alternative learning rules, challenges, and research directions.

Exploring how carefully curated, minimal datasets can unlock powerful LLM capabilities and reduce training costs.

Our AI systems generate high-entropy reinforcement learning data through real-world interactions, abstract and generalise via text diffusion models, and continuously retrain using no-propagation techniques. We're building a self-contained loop of autonomous improvement: this is artificial super-intelligence.

Why imitation will never yield true AGI—and how Superior Agents chart a radical new path.

Mainstream AI today is confined within a human-shaped box: the data we collect and the benchmarks we devise. These models master the art of mimicry—reproducing patterns we've labeled and approved—but they cannot transcend those patterns. If an AI were to discover a strategy that outperforms human reasoning, our evaluation frameworks would likely flag it as "incorrect," simply because it deviates from the training labels. The result is a self-limiting loop: intelligence capped at the level of its creators.

In this short article, we present a summary of current work on the grokking phenomenon that emerges when AI models are significantly over-trained. We suggest that this provides evidence of the model's attempts to define truth inductively through the creation of consensus sets within the base training set and encode it via patterns overlaid upon the same parameters used to memorize this set.

This system is currently under experimental training, and is periodically updated to reflect modifications made during the development process.