))
Diffusion LLMs: A Paradigm Shift in Text Generation
Fetch error
Hmmm there seems to be a problem fetching this series right now.
Last successful fetch was on December 04, 2025 13:34 (
What now? This series will be checked again in the next day. If you believe it should be working, please verify the publisher's feed link below is valid and includes actual episode links. You can contact support to request the feed be immediately fetched.
Manage episode 470371279 series 3605659
内容由Kabir提供。所有播客内容（包括剧集、图形和播客描述）均由 Kabir 或其播客平台合作伙伴直接上传和提供。如果您认为有人在未经您许可的情况下使用您的受版权保护的作品，您可以按照此处概述的流程进行操作https://zh.player.fm/legal。
In a groundbreaking development, Diffusion Large Language Models are revolutionizing the field by generating entire responses at once, using a technique inspired by text-to-image generation. This innovative approach, developed by Inception Labs, promises to be 10 times faster and 10 times less expensive than traditional autoregressive models that generate one token at a time. Unlike autoregressive models, diffusion models refine a rough, almost nonsensical text into a coherent solution through iterative steps. This leap in speed, achieving over a thousand tokens per second on standard NVIDIA H100 chips, drastically reduces waiting times and enables more test time compute. This breakthrough not only accelerates coding processes but also facilitates more advanced reasoning, error correction, and controllable generation, opening new possibilities for AI agents, edge applications, and various use cases. According to AI experts like Andrej Karpathy, this diffusion model may also unlock new unique psychology or new strengths and weaknesses, potentially leading to new behaviors in intelligent models.
Podcast:
https://kabir.buzzsprout.com
YouTube:
https://www.youtube.com/@kabirtechdives
Please subscribe and share.
325集单集
分享
