David Jurgens is now on BlueSky only
@david__jurgens
Associate Professor at @UMSI and @UMichCSE working in computational social science and NLP. PI of the Blablablab blablablab.si.umich.edu
ID: 221857665
http://jurgens.people.si.umich.edu/ 01-12-2010 20:21:16
373 Tweet
2,2K Takipรงi
566 Takip Edilen
๐ขIs current โhuman-AI alignmentโ research clarified and comprehensive? ๐ค We systematically reviewed 400+ papers across HCI, NLP, and ML to develop a framework for ๐ซ<>๐ค"Bidirectional Human-AI Alignment", encompassing the dual paths of โAligning AI to Humanโ and โAligning Human
Let's talk about about REJECTION in science and... geography๐ค New paper Hong Chen Chris Rider (chrisrider.bsky.social) David Jurgens is now on BlueSky only documents some important new facts: papers.ssrn.com/sol3/papers.cfโฆ (1/6)
How can we improve human and AI alignment? ๐ University of Michiganโs Dr David Jurgens is now on BlueSky only and Dr Hua Shenโจ will discuss how we can get AI to align better with human practices so that LLM models can work in more complex situationsโwith nuance. Register now ๐ resources.prolific.com/ai-alignment-wโฆ
๐ฉ๐ผโ๐ป Real or Robotic? ๐ค Can LLMs accurately simulate qualities of human responses in dialogue? Human conversations with LLMs are great for assessing the capabilities of LLMs. But having lots of folks chat with LLMs is challenging (๐ฐโณ๐ต๏ธ). Could we have another LLM *simulate*
๐๏ธ What if the way we prompt LLMs might actually hold it back? ๐จ Assigning personas like "helpful assistant" in system prompts might *not* be as helpful as we think! โจ Check out our work accepted to Findings of EMNLP 2025 โจ ๐ arxiv.org/abs/2311.10054 ๐งต [1/7]
![๐๏ธ What if the way we prompt LLMs might actually hold it back?
๐จ Assigning personas like "helpful assistant" in system prompts might *not* be as helpful as we think!
โจ Check out our work accepted to Findings of <a href="/emnlpmeeting/">EMNLP 2025</a> โจ
๐ arxiv.org/abs/2311.10054
๐งต [1/7]](https://pbs.twimg.com/media/GalxWUlagAALbNR.jpg "Mingqian Zheng (@elisazmq_zheng) on Twitter photo ๐๏ธ What if the way we prompt LLMs might actually hold it back?
๐จ Assigning personas like \"helpful assistant\" in system prompts might *not* be as helpful as we think!
โจ Check out our work accepted to Findings of <a href=\"/emnlpmeeting/\">EMNLP 2025</a> โจ
๐ arxiv.org/abs/2311.10054
๐งต [1/7]")
![[1/12] Optimizing prompts for specific tasks has been key to improving LLM performance, but what if we optimize prompts on system level to work well on *all* tasks? Check out ๐ฑSPRIG, a genetic system prompt optimizer that help unlock LLMs' full potential: arxiv.org/abs/2410.14826](https://pbs.twimg.com/media/Garrm1TXAAAFLVm.jpg "Lechen Zhang (@leczhang) on Twitter photo [1/12] Optimizing prompts for specific tasks has been key to improving LLM performance, but what if we optimize prompts on system level to work well on *all* tasks? Check out ๐ฑSPRIG, a genetic system prompt optimizer that help unlock LLMs' full potential: arxiv.org/abs/2410.14826")
This was a really fun collaboration and I'm super proud of Lechen Zhang for all his amazing work on SPRIG. He's applying for PhD programs this fall so if you're looking for a great student, keep an eye out!
Check out my poster on Trauma Event Detection tomorrow at 2pm (Jasmine) #EMNLP2024. /w Tobias Leemann David Jurgens is now on BlueSky only Jรผrgen Pfeffer ๐ @jurgenpfeffer.bsky.social Gjergji Kasneci aclanthology.org/2024.findings-โฆ
Excited to share our๐ฅNew #NeurIPS2024 Paper๐ฅ: ๐๐๐ง ๐๐๐ซ๐ ๐ ๐๐๐ง๐ ๐ฎ๐๐ ๐ ๐๐จ๐๐๐ฅ ๐๐ ๐๐ง๐ญ๐ฌ ๐๐ข๐ฆ๐ฎ๐ฅ๐๐ญ๐ ๐๐ฎ๐ฆ๐๐ง ๐๐ซ๐ฎ๐ฌ๐ญ ๐๐๐ก๐๐ฏ๐ข๐จ๐ซ? The project website (paper, code, demos): agent-trust.camel-ai.org The slides: tinyurl.com/mryyusmm ๐TLDR: We
