Will BDH replace Transformers? Post-Transformer AI Explained for Everyone

This AI thinks without words (no Chain of Thought!). Could it offer an alternative to today’s Transformer-based LLMs? In this video, we explore Pathway’s BDH ("Post-Transformer") architecture and BDH-CQ, the reasoning paper that topped Hugging Face’s August 2026 papers list. The makers of this model, Pathway, also shared separate pretraining experiments reaching 600 billion parameters, showing Transformer-like scaling. We start with the basics and explain latent reasoning, internal memory and ARC-AGI through simple examples. Can AI reason without a written Chain of Thought, and what could this mean for the future of LLMs? BDH-CQ paper: [https://arxiv.org/abs/2608.09888](https://arxiv.org/abs/2608.09888) HF Papers: [https://huggingface.co/papers/month/2026-08](https://huggingface.co/papers/month/2026-08) CHAPTERS: 00:00 Introducing BDH-CQ and the 600B scaling report 03:06 Chain of Thought and the cost of reasoning 05:32 The LLM memory problem and KV cache 06:45 Internal memory and continual learning 08:01 What Post-Transformer AI means 09:43 How the BDH architecture works 10:42 Latent reasoning: the chess player example 12:34 Why BDH is inspired by the brain 15:41 Four key ideas behind BDH 20:16 BDH vs BDH-CQ 23:39 ARC-AGI-1 explained with a puzzle 26:45 Benchmark costs and reasoning effort