🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

Good read: The Leaderboard Illusion: alphaxiv.org/abs/2504.20879 Big Tech commercially dependent on marketing model performance for revenues putting their best models out on Chatbot Arena is not surprising. I would argue against prohibiting score retraction after submission and

Are LLM leaderboards no longer trustworthy? cohere's deep dive reveals how Chatbot Arena scores can be gamed 🔍 Providers test 10–27 private models & submit the best 📊 Proprietary models receive 2–3× more data ⚠️ Rankings may reflect overfitting Trending on alphaXiv 📈

🚨Bright week for agents and representation learning, notably including X-Fusion’s remarkable progress in multimodal capabilities 🚀 Check out the top 10 papers for the week👇 - From LLM Reasoning to Autonomous AI Agents: A Comprehensive Review - Reinforcement Learning for

The most comprehensive survey (800+ papers) on LLM safety is here 👀 It highlights how agents, with tools, memory & env access, dramatically expand the attack surface This paper offers a "Full-Stack" analysis of these critical, often overlooked, risks Trending on alphaXiv 🚀

Trying out alphaXiv today. Loving how you can ask scientists questions within the context of their papers. The authors of the Absolute Zero paper (self-learning RL) @AndrewZ45732491 Shenzhi Wang🌟 and Qingyun Wu are kindly answering questions at alphaxiv.org/abs/2505.03335…

🚨Dive into Absolute Zero and RM-R1's major breakthrough in reasoning, unveiling another exciting week for AI.🚀 Check out the top 10 papers for the week👇 - Absolute Zero: Reinforced Self-play Reasoning with Zero Data - RM-R1: Reward Modeling as Reasoning - ZeroSearch:

alphaXiv is now my favorite way to keep up with arXiv. Thank you Raj Palleti et al. for building it!

Can reward models reason? RM-R1 frames reward modeling as a reasoning task—with structured rubrics, long-form evaluations, and verifiable justifications 🧠 +13.8% accuracy on benchmarks & increased interpretability 📊 32B RM-R1 beats GPT-4o & Claude Trending on alphaXiv 📈

"The S in MCP stands for Security" See this new review of MCP and its security considerations Key takeaways: ➡️ Security threats exist in ALL phases: creation, operation & update ➡️ Highlights risks of unofficial auto-installers & community servers

What would truly open-source AI look like? Not just open weights, open code/data, but *open development*, where the entire research and development process is public *and* anyone can contribute. We built Marin, an open lab, to fulfill this vision:

I just created Gaussian-Splatting community on alphaXiv for sharing the latest advances in this area. Please join the community for discussions 👏

Introducing Claude 4 Sonnet for understanding arXiv papers 🚀 Highlight any section of a paper to ask questions and “@” other papers to quickly add to context and compare results, benchmarks, etc.

🚨Prompt-to-A* Publication has been achieved 🤖🔬A fully AI-generated research paper has been accepted to the main conference of ACL 2025! Intology's research agent, Zochi, discovered and implemented a state-of-the-art jailbreaking attack on LLMs, which has been accepted to

This is pretty remarkable – AI systems learning to self-improve We're seeing a wave of research where AI isn't just learning from human feedback, it's starting to figure out how to improve itself using its own internal signals A subtle but profound shift.

"Can Large Reasoning Models Self-Train?" A brilliant paper from CMU showing LLMs can improve at math reasoning WITHOUT human labels - just learning from their own consistency. Early results rival models trained on ground-truth answers.

🚨There’s a new ceiling for efficient reasoning with the rise of Learning to Reason without External Rewards, along with AgriFM pushing the boundaries of AI to even agriculture🚀 Check out the top 10 papers for the week👇 - Paper2Poster: Towards Multimodal Poster Automation

Turns out you can jailbreak VLMs with memes A team of Korean researchers showed that pairing harmful prompts with everyday memes makes VLMs way more likely to generate dangerous content than text-only attacks across 50K+ examples

LLMs amaze at what they can do, LLMs amaze at what they can't do The dichotomy is as fascinating as it is frustrating

Introducing the Automated Research community on alphaXiv, created by Intology This Friday, we are hosting Samuel Schmidgall from DeepMind to discuss his work on automated research agents. Come chat with our team and the community! lu.ma/j9r9vf3m 🧵👇