Search papers, labs, and topics across Lattice.
2
0
4
Gemini Embedding 2's unified multimodal embeddings beat specialized models across diverse tasks and even generalize zero-shot to niche fields like astronomy and culinary arts.
Correcting a vision-language model's "hallucinations" is now as simple as pinpointing and editing the right intermediate representation, sidestepping costly retraining or dual inference.
