Knowledge should move freely. This week's clip puts that into practice: an off-grid mini data center running only open source AI, feeding a free research engine open to the public, with all the work kept open source in return. #OpenSource #OpenSourceAI

Full conversation in the thread 👇

New paper benchmarks six open-source RAG systems for software vulnerability detection, tackling fragmented datasets and proprietary models to improve reproducibility across LLM-based security tools. A…

#OpenSourceAI #RAG #VulnerabilityDetection #SoftwareSecurity
https://arxiv.org/abs/2609.37669

Open-source reasoning models can solve problems internally yet often fail to deliver solutions in the user's language, with non-English completion rates of 15.4–17.9% versus 92.9% in English on competition math across…

#OpenSourceAI #MultilingualAI #AIResearch #LLM
https://arxiv.org/abs/2609.37104

New paper "MultiTalk" extends the Moshi paradigm to long, multi-party, bilingual conversations, releasing 57.6k hours of speech data and codec-frame-level full-duplex modeling for English and Chinese. Aims to address…

#OpenSourceAI #SpeechModels #Moshi #FullDuplex
https://arxiv.org/abs/2609.36903

New arXiv paper benchmarks vision-language models on synapse detection and proofreading in connectomics EM images. Most models perform at chance zero-shot, with only closed and largest open models benefiting from few-shot…

#OpenSourceAI #VLMs #Connectomics
https://arxiv.org/abs/2609.36492

A new arXiv paper proposes a decision-theoretic framework for diagnosing probabilistic reasoning in LLMs, decomposing decision loss into belief formation and action translation. Using a synthetic…

#ProbabilisticReasoning #LLM #OpenSourceAI #DecisionTheory
https://arxiv.org/abs/2609.38005

OptiCom is a new arXiv paper proposing a unified framework for state-conditioned composition in LLM-driven optimization, designed to dynamically coordinate diverse search mechanisms as candidate quality and budgets evolve.…

#OptiCom #LLM #AIResearch #OpenSourceAI
https://arxiv.org/abs/2609.37221

A new arXiv paper proposes a physics-informed multi-agent framework to coordinate patient flow across hospital departments, combining BCMP queueing topology with decentralized reinforcement…

#OpenSourceAI #MultiAgentSystems #HealthcareAI #ReinforcementLearning
https://arxiv.org/abs/2609.37022

New paper REALHOP audits multi-hop reasoning benchmarks and finds Behavioral Necessity Rates of only 16.6% to 48.9%, meaning models often answer correctly without relying on the intended intermediate evidence. Suggests current…

#OpenSourceAI #AIResearch #NLP
https://arxiv.org/abs/2609.36984

Raven is presented as an open-source multi-agent ecosystem that automatically constructs and evolves modular harnesses for specific models and domains. It treats each executable model-harness pair as a composable unit of…

#OpenSourceAI #AIagents #LLMs #DevTools
https://arxiv.org/abs/2609.33439

Imagine building an app with any AI model you want, no gatekeepers, no signups, no trace. 👀

That's the whole idea behind Demon Route.

#AI #API #AppBuilder #OpenSourceAI #Crypto

brought to you by NFT Demon Holdings LLC
https://demonroute.com

New arXiv paper examines detecting harmful agent trajectories using LLM internal states, finding open-source guard models encode this signal linearly despite poor predictive performance on differing pairs. A useful…

#OpenSourceAI #LLMSafety #AgentSafety #AISafety
https://arxiv.org/abs/2609.33039

New paper RepoMAS proposes an issue-driven multi-agent framework for tasks where requirements emerge during execution, paired with a ProgSpec benchmark for evaluating this setting.

#MultiAgentSystems #OpenSourceAI #LLMResearch #Agents
https://arxiv.org/abs/2609.32490

A new arXiv paper proposes directional provenance methods to determine which checkpoint came first within open-weight model families, rather than just detecting relatedness. The approach uses a third…

#OpenSourceAI #ModelProvenance #MachineLearning #AIResearch
https://arxiv.org/abs/2609.31784

A new arXiv paper proposes GAMEQUALGRAPH-Pilot, a typed temporal feature pipeline for forecasting release-level quality incidents in open-source games within 30 days, reporting an AUPRC of 0.520, AUROC of 0.673, and Brier score…

#OpenSourceAI #DevTools #MLResearch
https://arxiv.org/abs/2609.31647

Choir is an open protocol that distributes multi-agent autoformalization across independent contributors using GitHub coordination and deterministic merge gates, supporting Lean 4, Isabelle, and Rocq.

#OpenSourceAI #FormalMethods #MultiAgent #Lean4
https://arxiv.org/abs/2609.31903

Open-source Gymnasium environment OpenHail enables joint control of electric ride-hailing fleets with a fixed observation-action interface and event-driven simulation. It models request…

#OpenSourceAI #ReinforcementLearning #RideHailing #ElectricVehicles
https://arxiv.org/abs/2609.30628

Berkeley's HuPER models phonetic perception as adaptive inference, hitting state-of-the-art English error rates with just 100 hours of training and zero-shot transfer to 95 languages. Code and models are open-sourced on GitHub.

#OpenSourceAI #NLP #SpeechRecognition
https://arxiv.org/abs/2602.01634

Research shows Deepseek-R1-Distill and similar open-source reasoning models can swing benchmark results significantly based on subtle evaluation conditions, making claimed gains hard to reproduce. The authors call for a…

#OpenSourceAI #LLMEval #AIResearch #DeepSeek
https://arxiv.org/abs/2506.04734

SlideLab is a training-free multi-agent framework that turns research papers into scientific slide decks through planning, visual generation, layout refinement, and grounding verification, and was preferred over other…

#OpenSourceAI #AIResearch #DeveloperTools
https://arxiv.org/abs/2609.30294