The Archive

Analysis of mismatch between automated animal identification method evaluation and real ecological data collection practices in conservation.

RSRCC: 126k remote sensing benchmark for localized change question-answering, built via retrieval-augmented best-of-N ranking.

unsloth releases GGUF quantized version of Qwen3.6-27B with embedded weights.

pAI/MSc: open-source multi-agent system reducing human steering to convert ML theory hypotheses into submission-ready manuscripts.

Study links calibration, curvature, and margins during neural network training; Expected Calibration Error tightly coupled to Hessian properties.

Meta employees' activity at work is now being used to train the company's AI agents. As reported by Reuters, Meta is installing a tool it calls Model Capability Initiative (MCI) on US-based employees' computers that runs in work-related apps and websites, recording mouse movements, clicks, keystrokes, and occasional screenshots. The data from this tool will be used to train the company's AI models to get better at interacting with computers the way humans do, including automating work tasks like those Meta's employees perform on the job. According to Reuters, the data from MCI won't be "used ...

Infosys said the integration will be used to help its clients modernize software development, automate workflows and deploy AI systems, initially focusing software engineering, legacy modernization, and DevOps.

Compares 6 discretization schemes (ZOH, FOH, BIL, POL, HOH, fourth-order) for Vision Mamba to improve temporal fidelity in dynamic scenes.

SuperIgor: instruction-following framework where LM generates/refines plans via RL feedback loop, reducing manual annotation needs.

User argues for $40–$60 mid-tier Claude subscription option between $20 Pro and $100 Max plans to address pricing gap.

SmartVector adds temporal, confidence, and relational metadata to embeddings to improve RAG accuracy on versioned queries from 58% baseline.

"Emily Hart" is a young, AI-created conservative woman who likes to take off her clothes.

Clustered federated learning with differential privacy and privacy-preserving initialization reduces data heterogeneity impact on convergence.

S4D state space models mapped to nonlinear oscillator networks, enabling interpretability of long-range dependency capture via wave-based encoding.

Multi-agent RL framework for distributed energy resources to coordinate P2P retail trading and wholesale market participation.

LLM agents in repeated Avalon deception games develop reputation dynamics and social memory across 188 games, studying emergent multi-round behavior.

Graph diagnostic framework evaluates assurance case structure and detects AI-generated vs. human-authored compliance documentation via link prediction.

At one point people thought of you as better than OpenAI and Google. We know AI companies are losing money. \- Just say, "We don't release Mythos because it'd be too expensive." \- Just say "We're going to increase the prices of Pro and Max because we're running out of money" ... all this under-the-radar marketing firm BS just means that you've decided to hemorrhage social capital as well as financial capital. Why would you want to do this?

ProactAgent enables lifelong learning agents to proactively retrieve past experience and skills during task interaction rather than passively.

Vine Denoising Copula amortizes vine-copula modeling via reusable bivariate denoising for tractable high-dimensional density estimation.

Logical connectives identified as high-entropy fragility points in LLM multi-step reasoning; intervention via selective path control improves stability.

Schema-guided reasoning framework for clinical CRF completion via two-stage decomposition handles sparse field prediction with strict output contracts.

LayerTracer framework analyzes hierarchical representations and robustness bottlenecks across diverse LLM architectures including Transformer, GateDeltaNet, and Mamba.

Theoretical analysis of Bayesian softmax-gated mixture-of-experts models with focus on posterior distribution asymptotic behavior.

Cross-lingual quality classification for multilingual pretraining uses embedding-space consistency to transfer filtering strategies from high-resource to low-resource languages.

Multi-agent iterative search framework for scientific idea generation combines combinatorial innovation theory with LLM-based planning to reduce repetitive outputs.

Conceptual critique of generative AI evaluation paradigms argues benchmarks constitute rather than measure models and calls for descriptive alternatives honoring pluralist contexts.

Evian framework provides fine-grained semantic auditing of vision-language model training data to identify logical fallacies and factual errors beyond coarse-grained quality scores.

Scoping review and survey of stuttered-speech research identifies gaps between current systems and end-user needs; recommends grounding speech tech in stakeholder experiences.

Alibaba releases Qwen3.6-27B, open-source 27B dense model with agentic coding surpassing larger models, Apache 2.0 licensed.