Podcast: Pathway's 150M-Parameter BDH Model Aims Beyond Transformers
bigdata · x · 2026-09-24
The Data Exchange podcast interviews Pathway CEO Zuzanna Stamirowska on "post-Transformer" AI:
- BDH model: only 150M parameters, reasons in latent space instead of chain-of-thought, maintains state over long horizons, and learns continually from experience.
- ARC-AGI and reasoning cost: discusses ARC-AGI results and dramatically lower inference costs — framed as "intelligence per watt."
- Continual learning: fast weights, learning from examples at inference time, catastrophic forgetting.
- Long-horizon agents: argues they may require a fundamentally different architecture from transformers.
- Deployment: corporate sovereign AI, one-rack models, open weights, and on-device AI.
More from Infra
- Nscale's $103B IPO backlog is a monetization ceiling, pair-trade idea circulates — menhguin · 2026-09-24
- Oracle sends force majeure notice over New Mexico data center, rattling AI compute supply chain — AIFlow_ML · 2026-09-24
- WSJ: Top 5 Firms to Spend $4.2T on AI CapEx by 2029, Largest US Infrastructure Build Ever — annbordetsky · 2026-09-24
- stable-diffusion.cpp runs SD, Flux, Wan and Z-Image diffusion models in pure C/C++ — leejet · 2026-09-24
- NVIDIA's Model-Optimizer unifies quantization, distillation, pruning and speculative decoding — NVIDIA · 2026-09-24
- How much memory for a 30B model? A quick precision-to-VRAM calculation — ashishllm · 2026-09-24