Lots of folks talking about scaling LLM inference over this last year Internally, I’ve been developing and using a library that makes this extremely easy, and I decided to open-source it Meet the decoding library: github.com/benlipkin/deco… 1/7

Really neat work by Yapei Chang et al. Han Xiao was really pumped about this at #emnlp2024

Can't wait for mine to come in the mail 🤞🏼

Really interesting generative (not-AI) art exhibit: aleksandrajovanic.com/line/

☠️Search is dead… or isn't it? 📢Call for Proposals: Second Search Futures Workshop ECIR2025! #IR is transforming with #AI & evolving research paradigms. Let's shape the future of search! searchfutures.github.io w/Charlie Clarke Adam Roegiest Ian Soboroff | [email protected] Johanne Zhaochun Ren

This reminds me that I forgot to post a picture of my copy when I was in SD! Will remedy soon

I open twitter, it’s all about deepseek, LinkedIn again deepseek, YouTube CNBC/BBC/Bloomberg gives me deepseek, my Chinese news app is packed with deepseek, my wife and friends are asking me deepseek on WeChat, my colleagues are asking me to generate synthetic data using

I got into photography in the past few years and people more experienced than I always repeat the refrain: "don't just look at it on a screen, print your work!" Really cool physical artifact of the great work being done Jina AI, thanks Bo Han Xiao for sending!

Rohan Jha We did not desk reject "the whole of RecSys submissions" In fact, in the papers that we kept and we sent out to review, recommender systems is still /by far/ the largest category. We looked at each individual paper to decide if it was on-topic for SIGIR. #SIGIR2025

Attending the faculty hiring lectures this and last spring at UT have been some of the most inspiring/though provoking experiences for figuring out how to define my own research agenda. (positive subtweet at Avi Schwarzschild 's talk today on memorization and reasoning by recurrence)

I applied LLMs for query expansion and we wrote this article: It sees to work out-of-the-box and generally boost the performance of embedding models. However, it requires more latency. Would be interesting to see more about this. jina.ai/news/query-exp…

Welcome back, Doc2query!

Why in the world can't Claude render LaTeX?

This looks like the GPU version of the GDM scaling book jax-ml.github.io/scaling-book/ So much great info in both!

Imagine if they mutter their chain of thought reasoning process as they complete chores.

Learning to Rewrite Negation Queries in Product Search Amazon introduces a query rewriting approach that bridges the vocabulary gap between negation expressions in queries and product text, improving search precision by 3.17% for queries with negations 📝amazon.science/publications/l…

Existing IR/RAG benchmarks are unrealistic: they’re often derived from easily retrievable topics, rather than grounded in solving real user problems. 🧵Introducing 𝐅𝐫𝐞𝐬𝐡𝐒𝐭𝐚𝐜𝐤, a challenging RAG benchmark on niche, recent topics. Work done during intern Databricks 🧱

if you're curious why the new o3 prefers ‑ [U+2011] to the more standard, visually identical - [U+002D] this is exactly why

🆕Please stop forcing Clippy on those who want Anton latent.space/p/clippy-v-ant… ChatGPT-4o's glazing lays open Clippy vs Anton: The two extremes of desires in AI post-training and product

Effective Inference-Free Retrieval for Learned Sparse Representations Franco Maria Nardini et al. introduce an inference-free retrieval approach that learns optimal token scores at training time, turning query encoding into a fast table lookup operation. 📝arxiv.org/abs/2505.01452