Jonathan Fürst

🚀 While I am already packing my bags for Vancouver ✈️, I want to share that we will present SM3-Text-to-Query (arxiv.org/abs/2411.05521) at the NeurIPS'24 Datasets and Benchmarks Track! Here's a quick teaser. 🧵

@shloka_2 Nice that you enjoyed our work :-)

Read the paper on VersionRAG. Learned a new perspective on handling changes across versions. Answering questions like 'What changed between v1 and v2?', which traditional RAG finds tricky. I think this idea could extend to any use case where numbers hold special meaning. #RAG

VersionRAG: Version-Aware Retrieval-Augmented Generation for Evolving Documents Proposes a framework that explicitly models document versioning through hierarchical graphs. 📝arxiv.org/abs/2510.08109 👨🏽‍💻github.com/danielhuwiler/…

3/ 🔧 Try it yourself: GitHub: github.com/gayecolakoglu/… 📄 Paper: arxiv.org/pdf/2502.18179

2/ Our OFAT (one-factor-at-a-time) method achieves near-optimal results with just ~2.8% of the computational cost of full factorial search. Result: 13-37 point F1 improvements over baseline configs 📈

🎉 Our paper "Problem Solved? Information Extraction Design Space for Layout-Rich Documents using LLMs" has been accepted to #EMNLP2025! Ever struggled with LLM-based PDF extraction in your org? This one's for you! 🧵👇

Want to design better Text-to-Query systems? SM3-Text-to-query is your new go-to benchmark, offering a comprehensive evaluation across different query languages. Read more from @JF87. #LLM #SQL towardsdatascience.com/can-llms-talk-…

Happening now. Q&A with Jeff Dean and team on Gemini and the new Gemini 2.0 Flash #NeurIPS2024

Introducing SM3-Text-to-query: The first dataset to evaluate cross-query language and cross-database models for Text-to-Query systems. @JF87 compares text-to-query capabilities of LLMs in his newest article. #LLM #SQL towardsdatascience.com/can-llms-talk-…

5/ More Details: Paper: arxiv.org/abs/2411.05521 Code & Data: github.com/jf87/SM3-Text-… See you in Vancouver! Happy to hear feedback and discuss how we can push Text-to-Query research forward. 🌟 #NeurIPS2024 #NeurIPS #AI #TextToQuery #MedicalAI #LLM @ZHAW @DataScienceCH

4/ Extensible Design: SM3-Text-to-Query can be easily extended for additional query languages, actual patient databases, or multilingual questions. Stay tuned for more! 😉

🚀 While I am already packing my bags for Vancouver ✈️, I want to share that we will present SM3-Text-to-Query (arxiv.org/abs/2411.05521) at the NeurIPS'24 Datasets and Benchmarks Track! Here's a quick teaser. 🧵

Pretty crazy that Google Gemini is only a team of 47 people. 2 engineers 16 PMs 29 senior global directors of DEI

@_richtig_falsch @TechDialectic Das kann nicht sein…

@TechDialectic Das ist echt

Remigration ist nicht möglich - weil wir nicht nur Fachkräfte brauchen, sondern auch Menschen ... .

@Tiefseher Und das bei zusätzlichem Bevölkerungszuwachs!

Das Frustrierende ist nicht, dass die deutsche Wirtschaft 2023 in die Rezession rutschte. Das Frustrierende ist, dass die deutsche Wirtschaft 2023 schrumpfte, während andere, vergleichbare Volkswirtschaften expandierten. DAS ist es, was der Ampel-Regierung nun auf die Füße fällt

One the biggest issues with this AI safety stuff is censorship You can’t ask ChatGPT anything remotely controversial or have it generate photos that are routinely posted on Insta to millions of likes ChatGPT is way more censored than an internet search!! 🤯🤯 Why? How does that make sense?

@emilymbender Mhm, we don’t really know that. This is just a press release by Microsoft. Like what are the exact rules and scope of this trial is not clear. In the end it is probably more marketing than anything else.

@JF87 This is not a case for "hey it might work" complacency. This is a direct attack on the effective delivery of healthcare in that region.

Attn computational linguists in Spain: you've got a job to do here. It appears your medical authorities need some education about what synthetic text machines are. And shame on Microsoft for pushing this tech.