Check out our new work, RARR: Attributed Text Generation via Post-hoc Research and Revision arxiv.org/abs/2210.08726 When applied to the output of LLMs on a diverse set of generation tasks, RARR significantly improves attribution while otherwise preserving the original input.

New open-source language model from Google AI: Flan-T5 🍮 Flan-T5 is instruction-finetuned on 1,800+ language tasks, leading to dramatically improved prompting and multi-step reasoning abilities. Public models: bit.ly/3sbNPDJ Paper: arxiv.org/abs/2210.11416

New from Google Research! We advance multi-vector neural retrieval with AligneR, which sparsely aligns query and doc tokens. AligneR can adapt alignment for new tasks using just 8 examples, advancing SOTA and 10x faster than prior multi-vector models. arxiv.org/abs/2211.01267

This is an *amazing* way to re-engineer the scoring mechanism of late interaction / ColBERT retrievers! Instead of gathering all vectors in each retrieved document, they approximate missing vector scores via an upper bound per query token—and modify the objective fn accordingly.

Check out our new paper, XTR. XTR is a new multi-vector retrieval model that achieves new sota AND is easy to use!

Introducing Conditional Adapters (CoDA) from Google Research! Adaptation methods (e.g. Adapter and LoRA) can finetune LMs with minimal parameter updates, but their inference remains expensive. CoDA makes LMs faster to use, and works for three modalities! arxiv.org/abs/2304.04947

We’re proud to announce that DeepMind and the Brain team from Google Research will become a new unit: 𝗚𝗼𝗼𝗴𝗹𝗲 𝗗𝗲𝗲𝗽𝗠𝗶𝗻𝗱. Together, we'll accelerate progress towards a world where AI can help solve the biggest challenges facing humanity. → dpmd.ai/google-deepmind

Just how good are commercially available embedding APIs for vector search? An effort led by Ehsan Kamalloo evaluated a few of them - OpenAI @CohereAI Aleph Alpha - on BEIR and MIRACL... Check out the results! arxiv.org/abs/2305.06300 - forthcoming #ACL2023 industry track paper

So excited to see our recent work launched on Google Bard (bard.google.com)! The “Goggle it” button double-checks claims made by LLM, provide relevant sources, and finds “hallucinated” information.

I'm in Singapore for EMNLP! I'll be presenting our work "Selectively Answering Ambiguous Questions." arxiv.org/abs/2305.14613 Our goal here was to try to decouple uncertainty about the question from uncertainty about the answer, using a selective question answering approach.

Gemini 1.5 Pro - A highly capable multimodal model with a 10M token context length Today we are releasing the first demonstrations of the capabilities of the Gemini 1.5 series, with the Gemini 1.5 Pro model. One of the key differentiators of this model is its incredibly long

I’m pleased to announce Gecko 🦎, a new text embedding model developed at Google DeepMind and now available on Google Cloud ! Gecko is powered by an LLM distillation recipe, and is one step towards our goal to bridge LLM and retrievers. Promptagator 🐊, Gecko🦎, what’s next?

Google presents Can Long-Context Language Models Subsume Retrieval, RAG, SQL, and More? Long-context LM: - Often rivals SotA retrieval and RAG systems - But still struggles with areas like compositional reasoning repo: github.com/google-deepmin… abs: arxiv.org/abs/2406.13121

"just put the corpus into the context"! Long context models can already match or beat various bespoke pipelines and infra in accuracy on non-trivial tasks! Hadn't expected this so soon, and honestly was hoping to milk RAG impact for a little longer 🤪

Very cool results from the Gemini Embedding model!