--- schema: 1 type: video title: 'Teaching LLMs to Speak Spotify — Yves Raimond & Jacqueline Wood, Spotify' description: >- One in four US Spotify Premium subscribers now use a single system every day, which Spotify calls the Large Taste Model. Yves Raimond, Spotify's SVP and GM of AI & Personalization, explains how Spotify turned its recommendation system "on its head" to be LLM-native. It went from curated playlists, to recommendations like Discover Weekly, to what Spotify calls generative personalization: a steerable DJ, prompted playlists, an editable taste profile and personal podcasts, across a catalog of more than 100 million tracks plus podcasts and audiobooks. Staff ML Engineer Jacqueline Wood then shows how the models are trained with NEO, Spotify's four-stage recipe. Semantic IDs are added to an open-weight LLM like Qwen. The new tokens are grounded while the backbone stays frozen, which keeps its language ability intact, where continued pre-training wiped it out. Then the model is instruction-tuned across many Spotify tasks, which even helped cold-start audiobook recommendations. She also covers decoding choices (98% of semantic IDs come out valid even without constrained decoding), and how grounding LLM judges with user profiles and behavior raised their agreement with humans by 91% on ambiguous queries. Speaker info: Jacqueline Wood, LinkedIn: https://www.linkedin.com/in/jacquelinewood Related links: Spotify: https://www.spotify.com Timestamps: 0:00 Intro: making LLMs speak Spotify 0:47 Spotify's scale: 760M users and 100M+ tracks 1:52 From curation to recommendations 2:37 Generative personalization 3:02 From guessing to reasoning, from black box to steerable 4:07 A Spotify DJ you can steer 4:37 Prompted playlists 5:32 Taste profile 6:22 Personal podcasts 6:52 The Large Taste Model 7:27 One in four US Premium subscribers use it daily 8:03 Jacqueline Wood: how the models are trained 8:18 Semantic IDs for Spotify's catalog 9:03 Natural-language podcast recommendations 9:33 NEO: a four-stage training recipe 10:02 Domain grounding with a frozen backbone 10:52 Capability induction: multitask tuning 11:32 Multitask gains and cold-start audiobooks 12:32 Why frozen grounding beats continued pre-training 14:02 Beam search vs. constrained decoding 15:02 One system in production 15:47 Breaking habitual listening in podcast discovery 16:11 Grounded LLM judges 18:16 Scaling evaluation sets with LLM judges 18:56 Summary --- Watch along with the corrected transcript, summary, timestamps and resources on this talk's official AIE page: https://ai.engineer/talks/2LRIAfng7eA-youtube-3fe36899d20ca6ca3ed9 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/channel/aidotengineer/explore/video/teaching-llms-to-speak-spotify-yves-raimond-jacquel-50b91540a8b2 markdown: >- https://liquidgold.cc/channel/aidotengineer/video/teaching-llms-to-speak-spotify-yves-raimond-jacquel-50b91540a8b2.md llms: null parent: 'https://liquidgold.cc/channel/aidotengineer.md' channel: aidotengineer updated: '2026-10-23T22:20:47.975Z' language: null children: [] videoId: 2LRIAfng7eA durationSeconds: 1180 publishedAt: '2026-09-25T14:30:25.000Z' searchChunks: - ordinal: 0 text: null startSeconds: 60 endSeconds: 117 watch: 'https://liquidgold.cc/watch/2LRIAfng7eA?t=60' heldReason: rights - ordinal: 1 text: null startSeconds: 129 endSeconds: 161 watch: 'https://liquidgold.cc/watch/2LRIAfng7eA?t=129' heldReason: rights - ordinal: 2 text: null startSeconds: 172 endSeconds: 183 watch: 'https://liquidgold.cc/watch/2LRIAfng7eA?t=172' heldReason: rights - ordinal: 3 text: null startSeconds: 195 endSeconds: 250 watch: 'https://liquidgold.cc/watch/2LRIAfng7eA?t=195' heldReason: rights - ordinal: 4 text: null startSeconds: 261 endSeconds: 283 watch: 'https://liquidgold.cc/watch/2LRIAfng7eA?t=261' heldReason: rights - ordinal: 5 text: null startSeconds: 296 endSeconds: 339 watch: 'https://liquidgold.cc/watch/2LRIAfng7eA?t=296' heldReason: rights - ordinal: 6 text: null startSeconds: 351 endSeconds: 385 watch: 'https://liquidgold.cc/watch/2LRIAfng7eA?t=351' heldReason: rights - ordinal: 7 text: null startSeconds: 397 endSeconds: 418 watch: 'https://liquidgold.cc/watch/2LRIAfng7eA?t=397' heldReason: rights - ordinal: 8 text: null startSeconds: 428 endSeconds: 449 watch: 'https://liquidgold.cc/watch/2LRIAfng7eA?t=428' heldReason: rights - ordinal: 9 text: null startSeconds: 460 endSeconds: 483 watch: 'https://liquidgold.cc/watch/2LRIAfng7eA?t=460' heldReason: rights - ordinal: 10 text: null startSeconds: 505 endSeconds: 516 watch: 'https://liquidgold.cc/watch/2LRIAfng7eA?t=505' heldReason: rights - ordinal: 11 text: null startSeconds: 563 endSeconds: 575 watch: 'https://liquidgold.cc/watch/2LRIAfng7eA?t=563' heldReason: rights - ordinal: 12 text: null startSeconds: 588 endSeconds: 610 watch: 'https://liquidgold.cc/watch/2LRIAfng7eA?t=588' heldReason: rights - ordinal: 13 text: null startSeconds: 622 endSeconds: 657 watch: 'https://liquidgold.cc/watch/2LRIAfng7eA?t=622' heldReason: rights - ordinal: 14 text: null startSeconds: 667 endSeconds: 704 watch: 'https://liquidgold.cc/watch/2LRIAfng7eA?t=667' heldReason: rights - ordinal: 15 text: null startSeconds: 714 endSeconds: 761 watch: 'https://liquidgold.cc/watch/2LRIAfng7eA?t=714' heldReason: rights - ordinal: 16 text: null startSeconds: 771 endSeconds: 851 watch: 'https://liquidgold.cc/watch/2LRIAfng7eA?t=771' heldReason: rights - ordinal: 17 text: null startSeconds: 863 endSeconds: 907 watch: 'https://liquidgold.cc/watch/2LRIAfng7eA?t=863' heldReason: rights - ordinal: 18 text: null startSeconds: 919 endSeconds: 953 watch: 'https://liquidgold.cc/watch/2LRIAfng7eA?t=919' heldReason: rights - ordinal: 19 text: null startSeconds: 964 endSeconds: 975 watch: 'https://liquidgold.cc/watch/2LRIAfng7eA?t=964' heldReason: rights - ordinal: 20 text: null startSeconds: 988 endSeconds: 1103 watch: 'https://liquidgold.cc/watch/2LRIAfng7eA?t=988' heldReason: rights - ordinal: 21 text: null startSeconds: 1115 endSeconds: 1138 watch: 'https://liquidgold.cc/watch/2LRIAfng7eA?t=1115' heldReason: rights --- # Teaching LLMs to Speak Spotify — Yves Raimond & Jacqueline Wood, Spotify One in four US Spotify Premium subscribers now use a single system every day, which Spotify calls the Large Taste Model. Yves Raimond, Spotify's SVP and GM of AI & Personalization, explains how Spotify turned its recommendation system "on its head" to be LLM-native. It went from curated playlists, to recommendations like Discover Weekly, to what Spotify calls generative personalization: a steerable DJ, prompted playlists, an editable taste profile and personal podcasts, across a catalog of more than 100 million tracks plus podcasts and audiobooks. Staff ML Engineer Jacqueline Wood then shows how the models are trained with NEO, Spotify's four-stage recipe. Semantic IDs are added to an open-weight LLM like Qwen. The new tokens are grounded while the backbone stays frozen, which keeps its language ability intact, where continued pre-training wiped it out. Then the model is instruction-tuned across many Spotify tasks, which even helped cold-start audiobook recommendations. She also covers decoding choices (98% of semantic IDs come out valid even without constrained decoding), and how grounding LLM judges with user profiles and behavior raised their agreement with humans by 91% on ambiguous queries. Speaker info: Jacqueline Wood, LinkedIn: https://www.linkedin.com/in/jacquelinewood Related links: Spotify: https://www.spotify.com Timestamps: 0:00 Intro: making LLMs speak Spotify 0:47 Spotify's scale: 760M users and 100M+ tracks 1:52 From curation to recommendations 2:37 Generative personalization 3:02 From guessing to reasoning, from black box to steerable 4:07 A Spotify DJ you can steer 4:37 Prompted playlists 5:32 Taste profile 6:22 Personal podcasts 6:52 The Large Taste Model 7:27 One in four US Premium subscribers use it daily 8:03 Jacqueline Wood: how the models are trained 8:18 Semantic IDs for Spotify's catalog 9:03 Natural-language podcast recommendations 9:33 NEO: a four-stage training recipe 10:02 Domain grounding with a frozen backbone 10:52 Capability induction: multitask tuning 11:32 Multitask gains and cold-start audiobooks 12:32 Why frozen grounding beats continued pre-training 14:02 Beam search vs. constrained decoding 15:02 One system in production 15:47 Breaking habitual listening in podcast discovery 16:11 Grounded LLM judges 18:16 Scaling evaluation sets with LLM judges 18:56 Summary --- Watch along with the corrected transcript, summary, timestamps and resources on this talk's official AIE page: https://ai.engineer/talks/2LRIAfng7eA-youtube-3fe36899d20ca6ca3ed9 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/channel/aidotengineer/explore/video/teaching-llms-to-speak-spotify-yves-raimond-jacquel-50b91540a8b2 · Channel: [AI Engineer](https://liquidgold.cc/channel/aidotengineer.md) > Search chunk text is held until public-source provenance is verified. ## Document chunks --- #### Chunk 0: Document excerpt (0:12 to 1:57) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 1: Document excerpt (1:57 to 3:03) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 2: Document excerpt (3:03 to 4:10) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 3: Document excerpt (4:10 to 4:43) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 4: Document excerpt (4:43 to 5:39) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 5: Document excerpt (5:39 to 6:25) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 6: Document excerpt (6:25 to 6:58) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 7: Document excerpt (6:58 to 8:03) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 8: Document excerpt (8:03 to 8:25) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 9: Document excerpt (8:25 to 9:35) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 10: Document excerpt (9:35 to 11:44) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 11: Document excerpt (11:44 to 12:41) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 12: Document excerpt (12:41 to 14:11) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 13: Document excerpt (14:11 to 15:07) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 14: Document excerpt (15:07 to 15:53) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 15: Document excerpt (15:53 to 16:15) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 16: Document excerpt (16:15 to 18:23) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 17: Document excerpt (18:23 to 18:58) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 18: Document excerpt (18:58 to 19:40) > Held: transcript rights do not permit this chunk's text.