Meta's Atomic Creator Tokens beat Semantic IDs: +19.8pt Recall@15 and 2.7x lower latency
_reachsumit · x · 2026-11-04
A Meta paper proposes Atomic Creator Token (ACT), replacing Semantic IDs in generative recommendation for cold-start creator recommendation.
The paper argues the SID layer, not the language model, is the binding constraint: the creator-to-SID map is many-to-one, so colliding creators are resolved by a popularity tie-break that caps recall and favors already-popular creators, while the frozen quantizer blocks gradients from reaching creator representations.
ACT gives each creator its own end-to-end-trained vocabulary token, making the mapping bijective and inference a direct top-k over creator logits. On production data with the token layer as the only variable, ACT improves creator-level Recall@15 by 19.8 points (1.84x), still leads with a 64x larger SID codebook, and serves at 2.7x lower per-user latency.
More from Research
- François Chollet: modeling brain wiring as an engineering problem constrained by genome size and time — CatAstro_Piyush · 2026-11-04
- OffsecCyber dataset combines HackerOne reports and CVEs to train pentesting LLMs — AccBalanced · 2026-11-04
- Monofact rate in training data governs LLM scientific discovery, preprint finds — CatAstro_Piyush · 2026-11-04
- qwenTPU compiles a TPU into Qwen3.5 weights, achieving exact integer matmul — CatAstro_Piyush · 2026-11-04
- Jev model improves ICL demonstration retrieval via distillation into a lightweight MLP — kalyan_kpl · 2026-11-04
- Indie researcher open-sources 10 years of math work: 146k files, AI-accelerated over 3 years — HamOnWhy · 2026-11-04