Six key papers on instruction-tuned embeddings: INSTRUCTOR, GritLM, FollowIR, Gemini Embedding and more
adhamelarabawy · x · 2026-10-12
The author supplements his argument on decision models and embedding disruption with a reading list: INSTRUCTOR (one instruction-finetuned embedder covering 330+ tasks with SOTA results at an order of magnitude fewer parameters), GritLM (unified generative and embedding representations), FollowIR, InF-Embed, Gemini Embedding, and Gemini Embedding 2. Together they trace the research line of pushing more intelligence into the query/retrieval interface.
Related event: Decision Models May Disrupt Embedding Models, Argues Viral Thread(2 posts)→
More from Research
- "Results Without Understanding" Is How LLMs Work—and Science Itself Is Changing — fkasummer · 2026-10-12
- The classic VAE tutorial that bridges deep learning and probabilistic modeling — goyal__pramod · 2026-10-12
- Anesthesia blocks consciousness via quantum microtubule effects, Stuart Hameroff argues ahead of Tucson conference — JosephJacks_ · 2026-10-12
- DeepMind's SynthID Bio protein watermarks now verifiable on Proteinbase — davidstutz92 · 2026-10-12
- Mathematicians push back on OpenAI's model-generated proofs — ZeeshanZiaML · 2026-10-12
- GaussiAnimate turns 4D captures into rigged 3D assets without physics simulation — janusch_patas · 2026-10-12