Neat... seems like cascade ranking has been rediscovered, Don Metzler ? dl.acm.org/doi/10.1145/20… Lag of roughly a dozen years... yup, sounds about right.

Terrible timing announcing this today... but bring your #LLaMA3 models over to the TREC 2024 RAG track to tackle the all important question: 'How often should you take your toddler to the potty when potty training?'

OpenAI, Cohere, Google, and now Snowflake... welcome to the embeddings game! Extra props for models released under Apache 2. snowflake.com/blog/introduci…

We are on a continual quest to simplify reproducibility. Anserini now allows you to reproduce runs with dense and sparse retrieval models (e.g., on MS MARCO and BEIR) directly from a fatjar, 'installed' via wget. Try it out, let us know what you think! anserini.io

🚀Thrilled to unveil our work in efficient zero-shot listwise reranking! LiT5 harnesses T5 models to challenge state-of-the-art standards, with significantly smaller models. Discover more: arxiv.org/abs/2312.16098

The obvious question: How do the latest prompt-decoder LLMs for listwise reranking perform on low-resource languages? For four African languages (Hausa, Somali, Swahili, and Yoruba), Mofe Adeyemi Orochimaru's Demeanour 🧜🏿‍♂️ Ronak Pradeep provide the answer: arxiv.org/abs/2312.16159

Prompt-decoder LLMs for listwise reranking too large for you? Introducing our new LiT5 family of listwise reranking models: nearly as good but *much* smaller. Yup, T5's still got tricks to offer! arxiv.org/abs/2312.16098

🌐 Ever wondered whether LLMs know what they don't know? Does your LLM confidently bullshit answers?

🌏 We introduce NoMIRACL to evaluate LLM robustness in RAG across 18 languages!

❌ GPT-4 can hallucinate answers with a high 33.2% hallucination rate!

📜arxiv.org/abs/2312.11361

💥💥 Catch us at #FIRE2023 starting tomorrow!
- Track Overview: 10:15am IST, Dec 15th
- Track Session: 11:00am IST, Dec 16th

We wrapped up CIRAL earlier, and we say a big thank you thank to participating teams 💯🎉🎉

Our public leaderboard is available: ciralproject.github.io/#timeline

We've been doing some research on scaling pretraining data and language models for African languages and I'm excited to share our research at EMNLP 2023!

Work done with Mofe Adeyemi Oreva Ahia J.O Abraham Owodunni David Ifeoluwa Adelani 🇳🇬 and Jimmy Lin

Here's a primer:

1 /

Dueling LLMs for reranking from my students... trying to remain neutral. Maybe let's just build RankZephyr-wo-GPT.

📢 We're excited to introduce RankZephyr, a fully open-source zero-shot listwise reranking LLM. It achieves state-of-the-art effectiveness competing the much larger RankGPT-4, all while not being exposed to human labels! Work done with @sahel_sharify and Jimmy Lin.
🧵[1/n]