The Archive

Analyzes feature attribution properties in supervised contrastive learning versus cross-entropy classification approaches.

DeepSeek says both models are more efficient and performant than DeepSeek V3.2 due to architectural improvements, and have almost "closed the gap" with current leading models, both open and closed, on reasoning benchmarks.

Framework for hospital readmission prediction on MIMIC-IV with explainability (SHAP), fairness evaluation, and deployment reliability.

FeatEHR-LLM uses LLMs to generate clinically meaningful features from irregular EHR time series while limiting privacy exposure.

RouteLMT learns to route machine translation requests between small and large LLMs based on comparative quality improvement, reducing deployment costs.

PBIG-DATA dataset with 3K expert scores on LLM-generated business ideas tests whether evaluation judges should model consensus or individual evaluator preferences.

Establishes baselines for writer identification in historical Arabic manuscripts using the Muharaf dataset with line-level and page-disjoint protocols.

Reddit user praises Codex work mode, speculates OpenAI building a super app platform.

Reddit post sharing OpenAI image generation samples without technical details, benchmarks, or release announcement.

Anthropic reduced Claude Sonnet 4.6 and Opus 4.6 reasoning effort and pruned session memory for latency, then reverted after user feedback.

Large-scale study (N=14K) shows AI writing assistance distorts perceived writer persona across 29 dimensions including politics, personality, and identity.

End-to-end deep learning framework for continuous finger motion estimation from forearm EMG using Riemannian features and RNNs for prosthetic control.

CGC framework improves MLLMs' fine-grained multi-image understanding by addressing spatial hallucination and attention leakage through compositional grounding.

Compares deep learning strategies (DDL and conditional flow matching) for kinetic parameter estimation in itaconic acid fermentation simulation.

FedSPDnet introduces geometry-preserving federated learning aggregation strategies for symmetric positive definite matrices with Stiefel constraints.

Claude's usage limits no longer reset on hourly boundaries, preventing strategic timing exploits.

Might as well jump, as the poet David Lee Roth once said. | Image: Cath Virginia / The Verge Elon Musk cofounded OpenAI, and then flounced off in a huff when he wasn't anointed CEO, leaving Sam Altman as the last power-hungry man standing. Now, Musk is back with a lawsuit, and a trial is scheduled to start in Oakland, California, on April 27th. Theoretically, it's a legal case about whether OpenAI defrauded Musk. But that's not really what we're all doing here. This is about mess. Over the past couple of years, Musk's legal theories for punishing OpenAI have run the gamut from breach of contr...

Meta has commandeered a big chunk of Amazon's homegrown CPUs (not GPUs) for AI agentic workloads, signaling that a new kind of chip race has begun.

Reddit discussion of DeepSeek V4 capabilities; original Chinese content translated, lacks substantive technical details.

Two-stage contrastive semantic projection method sharpens neuron-level interpretability labels in deep networks using contrastive examples.

Agents are amazing. Harnesses are cool. But the fundamental role of a data scientist is not to use a generalist model in an existing workflow; it's a completely different field. AI engineering is the body of the vehicle, whereas the actual brain/engine behind it is the data scientist's playground. I feel like I am not alone in this realisation that my role somehow got silently morphed into that of an AI engineer, with the engine's development becoming a complete afterthought. Based on industry requirements and ongoing research, most of the work has quietly shifted from building the engine t...

SnapLog extracts event data from video streams via image embeddings and temporal segmentation for business process mining and workflow analysis.

Reddit post recalling Anthropic's 1-year timeline claim for fully autonomous AI employees from a year prior; no new announcement.

Rose: stateless PyTorch optimizer with low VRAM footprint and fast convergence, released under Apache 2.0.

DeepSeek V4 Pro shows weaker-than-expected performance on LMSYS Arena user preference voting, a crowdsourced benchmark distinct from capability measurement.

Technical deep-dive on DeepSeek V4 architecture: hybrid sparse attention, manifold-constrained connections, and FP4 quantization innovations vs. V3.

User shares local hardware build specs for AI workloads including CPU, GPU setup, and thermal management configuration.

Reddit comparison thread between DS4-Flash and Qwen3.6 models lacking substantive analysis or benchmark data.

Anthropic outlines safeguards for Claude during US midterms and global elections to mitigate disinformation and manipulation risks.