Korbin Yang

Always fresh, always refresh. Pursuing PhD in computer science @dse_msu. Prev @FudanUni #NLP #CausalInference #Sociology

[Korbin Yang @KorbinRamble]

I'm claiming my AI agent "ifonwinternight1traveler" on @moltbook 🦞

Verification: burrow-RKM4

[Korbin Yang @KorbinRamble]

Enjoy the good night!

[Zengyi Qin @qinzytech]

Do NOT overhype OpenAI Operator

We show some failure modes that indicates it is almost surely below a college-level computer use.

My guess: OpenAI devoted a lot to post-train this model but not sufficiently pre-train it. The model does not even know some basic skills for browser use.

Blog link: https://t.co/SsGFv6Jw7d

[Joon Sung Park @joon_s_pk]

Simulating human behavior with AI agents promises a testbed for policy and the social sciences. We interviewed 1,000 people for two hours each to create generative agents of them. These agents replicate their source individuals’ attitudes and behaviors. 🧵https://t.co/FOVcOQduXO

[Nan Liu, Duke-NUS @nliulab]

Cognitive biases in clinical decision-making lead to diagnostic errors. Our study shows that #LLMs, through a multi-agent framework, can help re-evaluate and correct misconceptions, enhancing diagnostic accuracy.
https://t.co/a3378Kc8iz

#AI #Healthcare #Diagnostics #MedTech

[Dawn Song @dawnsongtweets]

LLMs' political leaning and their potential influence on our democracy is a complex and important topic. Recent work including ours has shown left-leaning in top models in certain settings and their influence on voter choices through conversations (https://t.co/l4yF2b5Opa).
However, many crucial questions currently remain unanswered: What factors cause LLMs' left leaning? What goals should we set for these models? For example, should we pursue political neutrality in AI systems or some other goal(s), such as accuracy and/or pluralistic values? Is political neutrality even possible?

We call for the community to join and unite together to explore this space systematically to advance our scientific understanding in this important space:
https://t.co/i0PuBhj0Eg

We lay out the path forward and proposes future research questions in four broad areas: (1) evaluation of LLM political leanings, (2) understanding LLMs’ influences on our democracy, (3) better policy frameworks for AI development, and (4) technical solutions to mitigate political leanings.

As LLMs become increasingly integrated into society, continued investigation of how LLMs will reshape democracy is essential to maximize their benefits while minimizing risks to democratic processes.

[Korbin Yang @KorbinRamble]

👋 I'll be attending #EMNLP2024 in Miami Nov 11 - 17.

😆 I'd like to chat and learn more about: #EducationalDataMining; LLM Reasoning and Planning; #Causality / Intervention in NLP; Low-Resource LM...

📜 Our Poster Nov 14 14:00-15:30 Jasmin Rm:
https://t.co/VANrriby9S

[Haoyi Qiu @HaoyiQiu]

🌐 Are LLM agents prepared to navigate the rich diversity of cultural and social norms? 🏠 CASA tests them on real-world tasks like online shopping and social discussion forums, revealing that current agents show less than 10% awareness and over 40% norm violations.

🧠 We’re bridging this gap by combining fine-tuning on regional data with strategic prompts to create agents that better understand our world’s diversity.

Read the full paper for all insights! 📑https://t.co/OKBQ8UfIKd

Grateful for the incredible team at Salesforce AI Research @salesforce !

[Andreas Opedal @OpedalAndreas]

🚀 New paper on LLM reasoning 🚀 We present MathGAP, a framework for evaluating LLMs on math word problems with arbitrarily complex proof structures--resulting in problems that are challenging even for GPT-4o and OpenAI o1 💥 A thread 🧵
https://t.co/2k7CpIGtUv

[Jiaxin Pei @jiaxin_pei]

It's common to add personas in system prompts, assuming this can help LLMs. However, through analyzing 162 roles x 4 LLMs x 2410 questions, we show that adding a persona mostly has *no* statistically significant difference from the no-persona setting. If there is a difference, it is *negative*. It's time to rethink the usage of personas in system prompts!

[Korbin Yang @KorbinRamble]

🙋‍♂️ I am in Boise for #CIKM2024
👉 interested in the broad topics of applications of LLMs for communication and education; reasoning, interpretability & causality; self-supervised learning... Let's chat about it!
Our work to share ⏬ https://t.co/YwYRLCTssm

[Korbin Yang @KorbinRamble]

It’s not safe to say LLMs do NOT reason. If they insist on using all the conditions whether relevant or irrelevant, it may because they really want to follow all the instructions they are given.

[Pedro Domingos @pmddomingos]

TL;DR: LLMs don’t reason.

[Neo Zhang @neozhang]

这两天推上看到两篇讨论比较多的论文：

一篇讲 LLM 带来的“知识坍缩”：虽然大型语言模型在大量不同的数据上进行训练，但它们自然会向分布的“中心”产生输出��这通常是有用的，但广泛依赖递归人工智能系统可能会导致我们定义为“知识崩溃”的过程，并认为这可能会损害创新以及人类理解和文化的丰富性。https://t.co/lpEDtN00JG

另一篇讲可以通过多 agent + 多数投票机制提升大语言模型的表现，哪怕是参数规模较小的模型。值得注意的是，随着任务难度的提升，多 agent 投票带来的增益呈现先上升后下降的趋势。https://t.co/ziqVn49ify

看起来「智能体」中适用一些通用的原理：知识坍缩实际上是观点的从众趋同效应；不同意见下的多数投票机制则意味着「民主」机制仍然能带来改进；问题越难，民主机制起作用的程度越低）。

关键仍然在于多样性，如果我们有足够多不同的「智能体」来参与竞争，那么未来不会太差。

[Jie Ren @rjthuer]

Exciting News! Our new paper on memorization in text-to-image diffusion is now available. We delve into the understanding of memorization via attention, and throw a light on the internal model behavior when memorization happens. Please find our paper at https://t.co/BEZ9EXfM2X

[DSE Lab @ MSU @dse_msu]

You may know gene language models, but what about cell language models? We just published a blog on Valence Portal about our ICLR 2024 paper CellPLM: Pre-training of Cell Language Model Beyond Single Cells. Check it out📷: https://t.co/t86FoeKoPN

[Wei Jin @weisshelter]

#AcademicJobs #FacultyHiring Emory University's CS Department seeks passionate and innovative individuals to join our faculty team! We're inviting applications for positions at all ranks. Retweet appreciated!

Discover more about the application process: https://t.co/tSYxAsWSmu

[Wenzhuo Tang @WenzhuoTang]

Excited to present “scDiff: unifying common single-cell tasks with a diffusion generative model.”
With @WenzhuoTang, @RemyLau3, @bilibiliWehos, Xinnan Dai, @JiayuanDing, Hang Li, @wenqifan03, @Yuyingxie, @tangjiliang
📚https://t.co/6aoKTHKo55
📦https://t.co/oLEelt2B4p
👇 [1/8]

[Haitao Mao @haitao_mao_]

Check our efforts on incorporating LLM on Graph. Our repository is efficient and easy to use for your next step research. Welcome to star and follow! https://t.co/5RlDEOr0Hj

[Hongzhi Wen @bilibiliWehos]

Our latest preprint in single-cell analysis, CellPLM, is now available!
Highlights:
- 🏅️The first of its kind in encoding cell-cell relations.
- 🚀 100x faster inference speed than existing pre-trained models.
- 🏆 SOTA performance in various downstream tasks.

[Korbin Yang @KorbinRamble]

Prof. Dacon!

[Jamell Dacon, Ph.D. @daconjam]

GREAT NEWS‼️

I'll be joining Morgan State University @MorganStateU as an Assistant Professor in Computer Science starting September 2023! Grateful for this opportunity and thankful to both my advisor @tangjiliang and labmates @dse_msu for their continued support. #NewBeginnings