TTT-Embed: Boosting Retrieval Up to 8.36 Without Weight Updates
_reachsumit · x · 2026-08-14
Google DeepMind introduces TTT-Embed, a framework that enhances dense retrievers via Test-Time Tuning of Embeddings.
Core Mechanism:
- Distills ranking rewards from a reranker or LLM judge into a lightweight vector.
- Adds this vector to frozen query embeddings without updating model weights or modifying the index.
Key Advantages:
- Applicable to closed-source models where parameter access is unavailable.
- Uses a single scope parameter to control reward reuse (global, task, or query level).
Evaluation:
Tested across 5 embedding models and 15 MTEB tasks, achieving up to +8.36 nDCG@10 improvement. The learned states also generalize effectively to unseen queries and tasks.
More from Research
- SIGCHI Announces Futures Summit in Paris, November 2026, to Reflect on Long-Term Future — plopesresearch · 2026-08-14
- Scott Domains and Semilattices: Generalizing via Linear Scott-Continuous Maps — jessi_cata · 2026-08-14
- 120-Page 'DeepSeek Harness Orange Book' Free and Open Source: AI-Written Framework Teardown — AlchainHust · 2026-08-14
- Meta/Oxford study: multimodal models need only 5% image-generation data, language training is key — rohanpaul_ai · 2026-08-14
- Exploring unified semantics for functional programming and CRDTs: domain-theoretic programming and semilattice homomorphisms — jessi_cata · 2026-08-14
- 3DGS training now takes just 12 seconds, setting new SOTA for MipNeRF360 — janusch_patas · 2026-08-14