--- schema: 1 type: video title: 'Text Diffusion — Brendan O’Donoghue, Google DeepMind' description: >- GPT-4o answered 40. Gemini 2.5 Flash answered 42 and stuck to it even after working through the reasoning incorrectly. The Gemini Diffusion model, considerably smaller than both, answered 60 on the first forward pass, then 49, then corrected itself to 39 once it finished reasoning. Bidirectional attention means it can see future tokens and go back to fix mistakes. Autoregressive models cannot do that. Brendon O'Donoghue covers why text diffusion is fast (24 denoising steps to generate 256 tokens means roughly 10x fewer memory transfers than autoregressive generation), what the tradeoff is (lower throughput at large batch sizes makes it expensive to serve at scale today), and what gets unlocked when latency drops to 2,000 tokens per second. The demos include a fake Wikipedia generated on the fly, a Reddit clone with AI generated comments and images, an operating system where every click generates the next screen, and a todo app built in 15 seconds by voice. Speaker info: - https://x.com/bodonoghue85 - https://bodono.github.io/ - https://www.linkedin.com/in/bodono/ Timestamps: 0:00 Introduction to Text Diffusion 1:02 How Text Diffusion Works (Training and Inference) 2:06 Gemini Diffusion Research Preview 3:04 Difference Between Autoregressive and Diffusion Models 4:02 Pros and Cons of Text Diffusion 6:13 Hardware Efficiency: Why Text Diffusion is Faster 8:47 Bidirectional Reasoning and Self-Correction 12:00 Dynamic and Adaptive Computation 14:26 In-place Text Editing 16:09 Low Latency Applications and Demos 20:05 Q&A Session --- Watch along with the corrected transcript, summary, timestamps and resources on this talk's official AIE page: https://ai.engineer/talks/r305-aQTaU0-text-diffusion-brendan-o-donoghue-google-deepmind Join our upcoming AI Engineer conferences (see https://ai.engineer landing page): - AIEi Shanghai, November 5–6 — the first for meeting top Chinese labs and engineers - AIE CODE in SF, November 10–12 — the top AI Coding conference returns - AIEi Sydney, December 7–8 — the first AIEi run alongside NeurIPS 2026 in Sydney --- canonical: 'https://liquidgold.cc/watch/r305-aQTaU0' markdown: 'https://liquidgold.cc/watch/r305-aQTaU0.md' llms: 'https://liquidgold.cc/watch/r305-aQTaU0/llms.txt' parent: 'https://liquidgold.cc/channel/aidotengineer.md' channel: aidotengineer updated: '2026-10-24T09:53:02.055Z' language: null children: [] videoId: r305-aQTaU0 durationSeconds: 1683 publishedAt: '2026-06-04T18:00:06.000Z' searchChunks: - ordinal: 0 text: null startSeconds: 26 endSeconds: 69 watch: 'https://liquidgold.cc/watch/r305-aQTaU0?t=26' heldReason: rights - ordinal: 1 text: null startSeconds: 80 endSeconds: 134 watch: 'https://liquidgold.cc/watch/r305-aQTaU0?t=80' heldReason: rights - ordinal: 2 text: null startSeconds: 134 endSeconds: 144 watch: 'https://liquidgold.cc/watch/r305-aQTaU0?t=134' heldReason: rights - ordinal: 3 text: null startSeconds: 144 endSeconds: 188 watch: 'https://liquidgold.cc/watch/r305-aQTaU0?t=144' heldReason: rights - ordinal: 4 text: null startSeconds: 199 endSeconds: 253 watch: 'https://liquidgold.cc/watch/r305-aQTaU0?t=199' heldReason: rights - ordinal: 5 text: null startSeconds: 253 endSeconds: 263 watch: 'https://liquidgold.cc/watch/r305-aQTaU0?t=253' heldReason: rights - ordinal: 6 text: null startSeconds: 263 endSeconds: 373 watch: 'https://liquidgold.cc/watch/r305-aQTaU0?t=263' heldReason: rights - ordinal: 7 text: null startSeconds: 263 endSeconds: 373 watch: 'https://liquidgold.cc/watch/r305-aQTaU0?t=263' heldReason: rights - ordinal: 8 text: null startSeconds: 383 endSeconds: 527 watch: 'https://liquidgold.cc/watch/r305-aQTaU0?t=383' heldReason: rights - ordinal: 9 text: null startSeconds: 383 endSeconds: 527 watch: 'https://liquidgold.cc/watch/r305-aQTaU0?t=383' heldReason: rights - ordinal: 10 text: null startSeconds: 383 endSeconds: 527 watch: 'https://liquidgold.cc/watch/r305-aQTaU0?t=383' heldReason: rights - ordinal: 11 text: null startSeconds: 539 endSeconds: 721 watch: 'https://liquidgold.cc/watch/r305-aQTaU0?t=539' heldReason: rights - ordinal: 12 text: null startSeconds: 539 endSeconds: 721 watch: 'https://liquidgold.cc/watch/r305-aQTaU0?t=539' heldReason: rights - ordinal: 13 text: null startSeconds: 732 endSeconds: 876 watch: 'https://liquidgold.cc/watch/r305-aQTaU0?t=732' heldReason: rights - ordinal: 14 text: null startSeconds: 732 endSeconds: 876 watch: 'https://liquidgold.cc/watch/r305-aQTaU0?t=732' heldReason: rights - ordinal: 15 text: null startSeconds: 876 endSeconds: 888 watch: 'https://liquidgold.cc/watch/r305-aQTaU0?t=876' heldReason: rights - ordinal: 16 text: null startSeconds: 888 endSeconds: 976 watch: 'https://liquidgold.cc/watch/r305-aQTaU0?t=888' heldReason: rights - ordinal: 17 text: null startSeconds: 986 endSeconds: 1205 watch: 'https://liquidgold.cc/watch/r305-aQTaU0?t=986' heldReason: rights - ordinal: 18 text: null startSeconds: 986 endSeconds: 1205 watch: 'https://liquidgold.cc/watch/r305-aQTaU0?t=986' heldReason: rights - ordinal: 19 text: null startSeconds: 1456 endSeconds: 1546 watch: 'https://liquidgold.cc/watch/r305-aQTaU0?t=1456' heldReason: rights - ordinal: 20 text: null startSeconds: 1456 endSeconds: 1546 watch: 'https://liquidgold.cc/watch/r305-aQTaU0?t=1456' heldReason: rights - ordinal: 21 text: null startSeconds: 1308 endSeconds: 1446 watch: 'https://liquidgold.cc/watch/r305-aQTaU0?t=1308' heldReason: rights --- # Text Diffusion — Brendan O’Donoghue, Google DeepMind GPT-4o answered 40. Gemini 2.5 Flash answered 42 and stuck to it even after working through the reasoning incorrectly. The Gemini Diffusion model, considerably smaller than both, answered 60 on the first forward pass, then 49, then corrected itself to 39 once it finished reasoning. Bidirectional attention means it can see future tokens and go back to fix mistakes. Autoregressive models cannot do that. Brendon O'Donoghue covers why text diffusion is fast (24 denoising steps to generate 256 tokens means roughly 10x fewer memory transfers than autoregressive generation), what the tradeoff is (lower throughput at large batch sizes makes it expensive to serve at scale today), and what gets unlocked when latency drops to 2,000 tokens per second. The demos include a fake Wikipedia generated on the fly, a Reddit clone with AI generated comments and images, an operating system where every click generates the next screen, and a todo app built in 15 seconds by voice. Speaker info: - https://x.com/bodonoghue85 - https://bodono.github.io/ - https://www.linkedin.com/in/bodono/ Timestamps: 0:00 Introduction to Text Diffusion 1:02 How Text Diffusion Works (Training and Inference) 2:06 Gemini Diffusion Research Preview 3:04 Difference Between Autoregressive and Diffusion Models 4:02 Pros and Cons of Text Diffusion 6:13 Hardware Efficiency: Why Text Diffusion is Faster 8:47 Bidirectional Reasoning and Self-Correction 12:00 Dynamic and Adaptive Computation 14:26 In-place Text Editing 16:09 Low Latency Applications and Demos 20:05 Q\&A Session --- Watch along with the corrected transcript, summary, timestamps and resources on this talk's official AIE page: https://ai.engineer/talks/r305-aQTaU0-text-diffusion-brendan-o-donoghue-google-deepmind Join our upcoming AI Engineer conferences (see https://ai.engineer landing page): - AIEi Shanghai, November 5–6 — the first for meeting top Chinese labs and engineers - AIE CODE in SF, November 10–12 — the top AI Coding conference returns - AIEi Sydney, December 7–8 — the first AIEi run alongside NeurIPS 2026 in Sydney --- Canonical: https://liquidgold.cc/watch/r305-aQTaU0 · Channel: [AI Engineer](https://liquidgold.cc/channel/aidotengineer.md) > Search chunk text is held until public-source provenance is verified. ## Document chunks > Missing: no stored document chunks are available.