Tune into the excellent discussion on our most actively discussed paper this week: To Code, or Not to Code! Here are the top 5 comments/questions! Thanks again @viraataryabumi and team! 1) How does pretraining on code data impact performance on non-English downstream tasks?
@askalphaxiv
-
alphaXiv Launches Google Scholar Integration Feature
By
–
alphaXiv now supports Google Scholar Integration!
— alphaXiv (@askalphaxiv) 21 septembre 2024
Add the link to your Scholar profile and we'll automatically sync all your arXiv papers. pic.twitter.com/IIjQXD3qRZalphaXiv now supports Google Scholar Integration! Add the link to your Scholar profile and we'll automatically sync all your arXiv papers.
-
Open Research Platform Hiring Full-Stack Backend Engineer
By
–
Join Our Team We're looking for a full-time full-stack/backend engineer to continue growing the platform and fulfill our mission of open & interactive research. If you have an excellent webdev background and align with our mission, please DM us or email contact@alphaxiv.org!
-
Direct Preference Optimization Authors Discuss DPO Impact
By
–
Thrilled to have the authors of Direct Preference Optimization answering questions on their paper! DPO offers a simple alternative to RLHF and has been hugely impactful, including being used to train Llama 3! Talk to the authors @archit_sharma97 and team directly to learn more!
-
Open Access Science and Democratic Peer Review Culture
By
–
Absolutely agree here. We hope that we can help by fostering a culture of open access within science, as well as making peer review a true community service that is democratic and accessible by everyone.
-
AI Scientist Platform Enables Research Reproducibility Testing
By
–
Sharing an idea that offers an exciting use case of the AI Scientist from @SakanaAILabs
! Rather than generating new ideas, one can adapt the code to pick a paper and test its results. "Reproducibility as a platform." Thanks lakshaytalkstocomputer and @cong_ml for the insight! -
TaskGen: Open-Source LLM Agentic Framework for Task Automation
By
–
Introducing TaskGen, an open-sourced task-based, memory-infused LLM agentic framework that uses an Agent to break down a task into subtasks and assigns each subtask to either an equipped function or another Agent. Talk directly with @johntanchongmin and the TaskGen team here!
-
Code Data Improves LLM Generalization Beyond Coding Tasks
By
–
New from @CohereForAI and @cohere
! To Code, or Not To Code? This paper reveals that including code data in LLM pre-training is critical for generalization far beyond coding tasks. The authors @viraataryabumi @ahmetustun89 and @sarahookr are here to answer your questions! -
Community Comments on Preprints Improve Transparency and Quality
By
–
We absolutely agree here and hope that community comments on preprints will help improve transparency and quality.
-
Community engagement beyond trending papers in AI research
By
–
Thanks for bringing this up @lauriewired
! At this early stage of the community, we agree that the most trending papers are receiving the most comments, but there are many papers that are not necessarily trending but still receive a question or comment from a curious student or