Search the full wire by company, model, lab, or keyword. Every story we have ever aggregated.
CrossMath benchmark reveals vision-language models rely predominantly on textual reasoning rather than vision grounding via controlled multimodal comparison.
HILBERT framework aligns long-sequence audio-text representations using cross-attentive aggregation and contrastive training in low-resource settings.
Gradient Fingerprint (GRIFT) detects and suppresses reward hacking in RL by identifying implicit model exploitation without examining surface-level reasoning.
BAGEL benchmark evaluates specialized animal knowledge in LLMs using bioRxiv, biotic interaction, audio, and Wikipedia sources with closed-book QA pairs.
Stratechery weekly digest covering infrastructure costs, satellite connectivity, and venture capital trends.
Kometo algorithm achieves optimal regret bounds for multi-fidelity optimization without requiring function smoothness or fidelity knowledge.
Probabilistic bias correction machine learning framework reduces systematic errors in subseasonal weather forecasts beyond 2-week horizon.
Benchmark compares token pruning compression across Qwen3, Gemma-3, Llama-3, Aya for Korean-centric NLP with English-Korean vocabulary optimization.
Two-stage deep learning framework combining object detection and behavioral analysis for exam cheating detection.
Latent Grammar Flow: neuro-symbolic generative framework for discovering differential equations from data using discrete flow models.
Participatory design process for reflexive governance of ACM FAccT conference combining in-person sessions, polling, and community input.
Optimal transport methods for quantifying domain shifts in geographic space to assess cross-region model adaptation.
Conformal prediction framework for LLMs using Layer-Wise Information scores from internal representations for robust uncertainty calibration.
Prices for "critical components" are surging because of massive data center investments.
AIFIND method for incremental face forgery detection using semantic anchors to prevent feature drift and catastrophic forgetting.
ChemGraph-XANES: agentic framework automating X-ray absorption spectroscopy simulation and analysis from natural-language task specifications.
RISE: scalable data attribution method for LLMs focusing on output-layer influence hotspots via sketching instead of full gradient indexing.
The gap between AI insiders and everyone else is widening, and the spending, suspicion, and even new vocabulary are starting to show it. While OpenAI is busy buying up everything from finance apps to talk shows, a certain shoe company just rebranded as an AI infrastructure play, and Anthropic unveiled a model it says is too powerful to release publicly …but apparently not too […]
Conditional GANs with LSTM for generating synthetic cryptocurrency price time series to address privacy and access restrictions.
Dairy Queen is becoming the latest fast food chain to get in on AI, as it's bringing a chatbot to dozens of its drive-thrus across the US and Canada. It aims to help speed up drive-thru service and "encourage customers to add more food to their orders," according to The Wall Street Journal. Following a test last year, Dairy Queen is widely launching the tech built by Presto, an AI company that already works with chains like Carl's Jr., Hardee's, Taco John's, and Fazoli's. In 2023, a report from Bloomberg revealed that Presto's AI drive-thrus may actually be assisted by human workers based in ...
MARCH: multi-agent framework emulating radiology department hierarchy for collaborative 3D CT report generation with reduced hallucinations.
JumpLoRA uses JumpReLU gating to induce sparsity in LoRA adapters, enabling dynamic parameter isolation for continual learning without catastrophic forgetting.
AtManRL trains differentiable attention masks via RL to improve faithfulness of chain-of-thought reasoning in LLMs by identifying decision-critical tokens.
SWNet applies Pyramid Vision Transformer and bimodal fusion to detect camouflaged weeds using visible and near-infrared imagery in agricultural settings.
Study unifies implicit reward and nudging test-time alignment methods as rejection sampling from graphical models, comparing rejection criteria for LLM alignment.
Paper predicts distributed training time accounting for mixed-precision settings, showing 2.4x variation and significant error in precision-agnostic models.
Dataset and XGBoost model for sentiment analysis of German Sign Language fairy tales using LLM-annotated text and MediaPipe-extracted video features.
Study evaluates LLM performance on automated qualitative coding of firearm violence survivor interviews, examining both capability gains and ethical risks.
The company says Claude Design is intended to help people like founders and product managers without a design background share their ideas more easily.