alphaXiv

Uno nunca sabe el alcance que puede tener una investigación científica. Nuestro último artículo, publicado el 14 de marzo, ya suma +15.000 visualizaciones, +300 likes y está en la posición no16 global en alphaXiv. Nada menor para una plataforma que frecuentan investigadores. Creo que el conocimiento en algoritmos de optimización no debería estar limitado solo a especialistas. En el futuro, las máquinas harán mejor la implementación que nosotros ??, mientras que los humanos nos encargaremos de lo realmente importante: idear, verificar y definir qué adaptar. ??? Dicho esto, aprovecho la oportunidad para dar 5 consejos que he aprendido en estos 4 a?os de PhD para escribir papers atractivos ????: 1?? Si una imagen puede condensar toda una investigación, ?a?ádela! Yo soy un pensador visual, así que para mí cada imagen es parte de la historia, desde la idea inicial hasta los resultados. No todos los investigadores son buenos dise?ando, pero si lo eres, ?aprovéchalo! Así, un revisor podrá decir que tus resultados no son relevantes o que el problema no es novedoso, pero nunca que no entendió tu trabajo. 2?? Para terminar rápido un paper, primero hay que tener listas todas las tablas e imágenes, entonces te sientas a escribir. ???? Como dije en (1), las imágenes marcan el camino del texto. 3?? La introducción es la parte más difícil del paper. ?? De ella depende que alguien decida leerlo completo o no. Por eso, es mejor escribirla al final, cuando ya tienes una visión global del trabajo. 4?? El título debe ser atractivo, pero no un clickbait. ?? La diferencia es clara: en el clickbait, lo único atractivo es el título; luego llega la decepción. 5?? El repositorio debe tener instrucciones claras para reproducir los resultados. Aunque no siempre es posible garantizar la reproducibilidad (sobre todo cuando se trabaja con LLMs), la intención debe ser facilitar al máximo la comprensión y replicación del trabajo. Eso; y para los curiosos, dejo el enlace a mi paper: https://lnkd.in/ev5gGzGt Cualquier pregunta me la pueden dejar en lo comentarios ??

????Today we are debuting Zochi, the world’s first Artificial Scientist with state-of-the-art contributions accepted in ICLR 2025 workshops. Unlike existing systems, Zochi autonomously tackles some of the most challenging problems in AI, producing novel contributions in days—from idea to finalized publication. With a standardized automated reviewer, Zochi’s papers score an average of 7.67 compared to other publicly available papers generated by AI systems that score between 3 and 4. ?? Read our full technical report here → https://lnkd.in/gKMkbuXe Zochi’s work covers diverse and open-ended problems at the frontier of AI → Papers & Code: https://lnkd.in/gT2QasWx → Discussions on alphaXiv 1?? CS-ReFT finetunes LLMs at the subspace level, enabling the much smaller Llama-2-7B to surpass GPT-3.5's performance using only 0.0098% of model parameters. https://lnkd.in/g7EWmcks 2?? Siege identifies critical vulnerabilities in language model safety measures, achieving a 100% success rate against leading LLMs. https://lnkd.in/gBrCgf6G 3?? EGNN-Fusion is an efficient architecture for protein-nucleic acid binding site prediction that reduces parameter count by 95%. (TBA, completed post ICLR workshop deadlines) Zochi’s papers received unanimously positive reviews: 1?? CS-ReFT received strong peer review scores (6,7,6), with reviewers commending its "clever idea" and effectiveness in addressing "a critical limitation of ReFT." 2???Reviewers gave Siege scores of (7,7), highlighting the paper's "effective, intuitive method" that is "significantly more effective than prior methods, necessitating a reassessment of existing AI defense strategies." In an exploratory exercise on MLE-Bench Kaggle competitions, Zochi achieves state-of-the-art results without any task-specific optimization, surpassing median human performance on 80% of tasks and securing medals on 50% of them. We're committed to maintaining scientific integrity. Since AI systems cannot take responsibility for their work, we do not believe they should be listed as authors and discourage the submission of AI-produced work without thorough human verification. We are in discussion with the workshop organizers of Zochi's accepted papers. If they approve, we would be honored to present this work and ensure Zochi's valuable contributions reach the research community. At Intology, we're building Artificial Scientists because we believe they are the ultimate application of artificial intelligence. By accelerating the pace of discovery across all fields, we can help address humanity's most significant challenges. Learn more: https://lnkd.in/g2AazfpK We are a small team, join us: [email protected]

AI Agents ?? Tackling Vaccine Hesitancy: A Glimpse into the Future of Public Policy Making ??? Just stumbled upon a fascinating research paper about the power of AI in understanding (and potentially influencing!) human behavior. The paper, "Can A Society of Generative Agents Simulate Human Behavior and Inform Public Health Policy? A Case Study on Vaccine Hesitancy" (link in comments!), explores whether AI agents can accurately model complex social phenomena like vaccine hesitancy and inform public health policy decisions. The researchers simulated public reaction to three different vaccination policies: 1.???? Financial incentives: Cash rewards for vaccination. 2.???? Ambassador programs: Engaging trusted community figures to promote vaccination. 3.???? Vaccine mandates: Requirements for vaccination to access certain services. The simulation was tested across several state-of-the-art LLMs, including: Llama-3-8B-Instruct Llama-3-8B-Instruct-Abliterated (with safety guardrails removed) Llama-3.1-8B-Instruct Qwen-2.5-7B-Instruct GPT-4o (used primarily for evaluation) It's mind-blowing to see AI being used to model such nuanced and critical societal issues! Big shoutout to Ajay Shenoy for introducing me to alphaXiv! It simplifies and summarizes research papers, making them super accessible. What's even more interesting is that, fueled by curiosity, I actually went back and skimmed the original paper (something I rarely do!). It really enhances the learning experience. #AI #PublicHealth #VaccineHesitancy #GenerativeAI #Research #Innovation #PolicyMaking #DataScience #ArtificialIntelligence

?? Read your favourite Arxiv papers as blogs! Simply replace "arxiv" in the url to "alphaxiv" If you find going through research papers too tedious, or need to quickly skim through the important results, then this is a great solution! By Rehaan Ahmad and Raj Palleti

Understanding research papers is much harder than it needs to be. Introducing blog-style overviews for arXiv papers. We used the Mistral OCR API and Claude Sonnet 3.7 to create beautiful research blogs with figures, key insights, and clear explanations from the paper Understand papers in minutes - not hours To use go to any arXiv URL and change the 'arxiv' to 'alphaXiv' and click 'blog'. ex: https://lnkd.in/gxEu2xV9 -> https://lnkd.in/g-w37GvV

Dear friends, colleagues, peers, Last week, the paper: "Non-Cooperative Games with Uncertainty" arXiv:2503.01889 has been released on arXiv, and now?I would like to share some additional resources to the paper: You can read and comment on the paper on alphaXiv (which also provides a decent AI-generated Overview): https://lnkd.in/dmmnN_Gy I made a GitHub repository for the project where I plan to collect additional resources (such as Talks, Software, etc.): https://lnkd.in/dCmb5rtA I wrote a little package in Wolfram Mathematica by which the Extended Equilibrium can be calculated for minimal games with uncertainty (such as the "Generals and the Weather" game in the paper.) https://lnkd.in/diqrUps5 For the previous announcing post on LinkedIn, follow this link: https://lnkd.in/dwfGQZF5 Thank you for reading, I hope you will find the paper, software etc., useful or interesting :) Feel free to give feedback or ask questions on alphaXiv, here or via email.

?? ?? ?? AlphaXiv is now my go-to for reading papers. Why? It really improves my ability to perceive research: 1?? ?? State-of-the-art LLMs at my fingertips—ready to answer questions about any part of a paper. 2?? ?? A seamless note-taking interface—essential for tracking progress and insights while reading. 3?? ?? Community comments—less valuable for now, but could become more useful with wider adoption. ???♀??? How to use it? Simple: replace “arxiv” with “alphaxiv” in any ArXiv paper URL—and you’re in!

Microsoft has claimed a breakthrough in quantum computing … but not everyone is convinced. A deeper look reveals that two expert referees strongly advised against publication, raising concerns about the evidence. The editors note: “The results do not represent evidence of Majorana zero modes.”

Nuestro preprint?"Improving Existing Optimization Algorithms with LLMs" publicado el pasado 12 de febrero?ya supera las?5800 vistas?y?100 likes?en?alphaXiv, posicionándose en el?top global?de la plataforma ?? En este trabajo presentamos una nueva metodología para usar los LLMs en el área de dise?o de algoritmos de optimización. Este enfoque no solo complementa las técnicas desarrolladas el a?o pasado, sino que también abre nuevas vías para la?mejora de algoritmos existentes. Lo pueden leer aquí: https://lnkd.in/eccbvsSk

1997: Deep Blue defeats Kasparov at chess 2016: AlphaGo masters the game of Go 2025: Stanford researchers crack Among Us Trending on alphaXiv ?? Remarkable new work trains LLMs to master strategic social deduction through multi-agent RL, doubling win rates over standard RL. Learn more and join the discussion with the authors here: https://lnkd.in/gYDveqv7