Tag
The article discusses a framework for AI progression from model-driven to data-driven to decision-driven, questioning whether the bottleneck shifts to decision-making as models and data become abundant in the post-training context.
Uranus is a next-generation simulation infrastructure for embodied AI that uses a joint-trajectory-conditioned autoregressive diffusion model to enable scalable, low-latency generation of robot simulations with streaming rollout and extensible control.
Sherpa is an advanced fiction writing AI trained on 5.5 billion minutes of data, designed to solve long-horizon consistency in storytelling, outperforming existing models like ChatGPT and Claude.
A website that allows users to randomly select and experience a life from all humans who have ever lived, using real historical data.
A Unified Particle Filter LSTM framework is proposed for data-driven process simulation, maintaining weighted recurrent-state hypotheses to handle partial observations in event logs, and outperforming baselines in reproducing routing and duration behaviors.
This article discusses how data has shaped neural architecture discovery and explores how automation could reshape future technological advancements.
The article introduces 'memoryfields', a portable file format for AI agent memory that uses Markdown files and optional SQLite indexes, advocating for a simpler, data-driven approach over existing complex systems.
This paper presents an LLM-driven approach using LLaMA 3 and a performance database to generate dynamic algorithm dispatch heuristics in linear algebra, with a case study on LU factorization to optimize computational performance.
MorphoGP is a Gaussian process framework for predicting equilibrium beach profiles under tidal influence, achieving improved performance over baseline models through category-specific experts.
This paper establishes tight generalization bounds for multi-dimensional hyperparameter tuning in data-driven algorithm design, using real algebraic geometry and a multi-regime lower-bound framework to resolve theoretical gaps.
This paper systematically compares word order preferences in decoder-only language models across artificial and natural languages, finding that preferences are data-driven and can reduce linguistic diversity with the widespread adoption of LLMs.
This paper benchmarks popular machine learning-based frameworks for hyperelastic constitutive modeling using the classic Treloar dataset, comparing their performance, cost, and trade-offs to provide practical guidance for researchers.
Mark Zuckerberg shares his product development philosophy, emphasizing using data and qualitative feedback to identify problems, then testing intuitive solutions as hypotheses, using the creation of News Feed as a case study.
Art museums are increasingly using data and analytics to inform curation and improve visitor experiences, moving from ivory towers to participatory centers.
This paper presents a video codec that stores video and audio as weights of a sinusoidal representation network, using knowledge distillation and quantization for compression. Experiments show a 2.61x compression ratio compared to the original network, but quality lags behind standard codecs like H.264 and HEVC.
Introduces Cluster-Weighted EDMD, a data-driven method that jointly learns a partition and per-cluster Koopman operators via expectation-maximization, improving prediction accuracy over standard EDMD on classical dynamical systems.
Presents a Variational Graph Autoencoder (VGAE) for detecting power flow solution feasibility in electric power networks, using the IEEE 118-bus case. The method distinguishes between problem infeasibility and algorithm non-convergence.
This paper proposes a novel data-driven pipeline using LLMs to extract semantic triples from unstructured text and construct knowledge graphs for monitoring technological convergence. The method is validated on arXiv preprints and USPTO patent applications, demonstrating scalability for technology forecasting.
This guide is based on 1765 real job descriptions and actual interview experiences, providing data-driven learning paths, interview preparation, and skill maps for those who want to switch to AI engineering. It is a practical open-source resource.
Someone connected Blender to Fable 5, and in just 20 minutes, it reconstructed the entire New York City building complex based on public building data, with the model to scale. The poster believes this 'check data before acting' approach is smarter than Opus 4.8, reflecting that AI is beginning to have the ability to 'do its homework'.