RT @pgasawa: Instruct-tuned models are getting better at following instructions and ‘reasoning’ every day, but they’re shockingly poor at g…

RT @BikakisNikos: 𝗔𝗖𝗠 @SIGMOD 𝗕𝗹𝗼𝗴 𝗣𝗼𝘀𝘁 (, where @sihemameryahia @leibatt @dryifanhu @domoritz @adityagp discuss th…

RT @shao__meng: DocWrangler - built on top of DocETL, streamlines LLM data processing through interactive development and intelligent assis…

RT @sh_reya: Introducing 📜DocWrangler: an open-source IDE for AI-powered data processing with built-in prompt engineering guidance and outp…

This is a neat workshop on a timely topic run by some terrific folks - take a look!

RT @ajay_frontiers: Absolutely agree with all the DocETL testimonials. Has been fun with my personal and useful projects (3 and counting!…

RT @sh_reya: I always feel so thrilled to hear such DocETL testimonials. We are still in the “day zero” of the project; working tirelessly…

RT @sh_reya: We'll be presenting our MLOps paper at CSCW today! Session 3e: Trusting the Machine (Guanacaste 1); 2:30pm Costa Rica/12:30pm…

RT @MadelonHulsebos: Tables are a goldmine for accurate, fresh, domain data that LLMs should be grounded in for RAG, factver/QA, text2sql:…

RT @UCBIDS: 🎦 Watch the recording of Madelon Hulsebos's seminar at BIDS! "There is actually a lot of structured data available on the web..…

Open-rank search (unusual for EECS) - please RT and apply!

RT @GromitC: Organizing LLM calls in MapReduce fashion is extremely promising (not only for document processing)!

Thanks for having me! It was fun to reminisce about the CIDR’11 paper and talk about how we’re revisiting those ideas in an LLM context (CIDR’24 and beyond!)

RT @aipengineer: Today, QuackChat brings you: - #DocETL Framework: @sh_reya from UC Berkeley's EPIC lab releases a new approach to documen…

RT @Smol_AI: [21 Oct 2024] Congrats to @sh_reya et al for finally publishing the DocETL paper! lots of great ideas in here, now with a det…

RT @GregKamradt: Shreya and team have put together a really unique way to think about long document processing as someone who's really int…

Excited that our DocETL preprint, led by @sh_reya - is out!! I'm particularly stoked about our 13 new rewrite "directives"—like rewrite rules, but more abstract—all entirely instantiated by LLM agents to rewrite pipelines into alternatives. All of these alternatives are evaluated by LLMs as well. To me, this work represents a great example of combining the best of what the DB and AI communities has to offer: declarative query specification + a methodical exploration of the search space to find performant (here mostly accuracy) query plans, coupled with LLM agents to instantiate and evaluate the query plans. Take a look and let us know what you think!!

RT @subZero_saj: 📢Excited to bring the DAIS workshop to ICDE'25 (w/ @SainyamGalhotra @FarihaAnna @MikeCafarella @sairamgv) The focus is on…

RT @MadelonHulsebos: Interested in making representation learning and generative models work for structured data (e.g. tables in relational…

One important direction we’re taking DocETL: more human input and feedback throughout the pipeline in a no-code interface. Take a look at our early ideas on this!