Use this view to inspect the underlying evidence corpus. For ranked developments, decision posture and interpretation, use Signals.
Raw feed or Signals?
Raw feed is chronological evidence. Signals ranks and interprets material change.
STAIF proposes a stage-wise optimization framework for LLMs to better follow complex instructions with multiple constraints, addressing limitations of holistic reward signals.
Research explores using schema-derived constrained decoding for MLIR code generation, improving generalization across dialects without retraining.
Research identifies 'Persona Collapse' in LLMs, where distinct agents converge into homogeneous behavior, limiting diversity in multi-agent simulations.
Research challenges the assumption that parameter-efficient fine-tuning (PEFT) methods like LoRA ensure memory efficiency in LLMs due to intermediate tensor scaling.
Researchers propose a latent-space language steering method using PCA to reduce unintended code-switching in multilingual LLMs during inference.
Anthropic is pitching US lawmakers on draft AI safety legislation based on its Advanced AI Framework, according to a Politico report.
DeepSeek announced DeepSeek-V4.1-Flash-1017, claiming faster completion speeds and shorter outputs via self-feedback reinforcement learning.
World announced World ID for banking and finance alongside AgentKit in beta to verify human authorship behind agentic money transfers.
Anthropic is reportedly in discussions with law firms to help train AI agents in negotiation tactics, according to the Financial Times.
World launched a suite of digital identity tools for banks to counter deepfake fraud and AI impersonation.
Hugging Face and AWS highlighted updates to TorchNeuron, simplifying the deployment of Hugging Face models on AWS Trainium chips.
The 25th BIS Annual Conference convened central bank governors and academics in Basel to discuss macro-financial stability.
McKinsey outlines four strategic shifts, including artificial intelligence and product innovation, reshaping the US retirement industry.
Anthropic plans to allocate 5% of its shares to a new fund aimed at addressing the public good and mitigating AI's labor market impact.
nCino and Parlay integrated AI intake to automate SBA loan processing, claiming over $500m in inquiries handled with 80% less manual work.
Anthropic's Emmanuel Ameisen discusses mechanistic interpretability, model internal planning, consciousness, and AI safety mechanisms.
An OpenAI engineer was arrested for allegedly placing a covert recording device in an executive conference room to spy on sensitive AI work.
Ant International has detailed an Agentic Mobile Protocol designed to orchestrate AI agents for mobile payments and financial tasks.
ClearToken and Consensys have partnered to provide 24/7 delivery-versus-payment settlement infrastructure for banks in the UK.
Microsoft pulled out of a UK data center deal with neocloud firm Nscale, impacting Nscale's initial public offering plans.
The $10tn US AI infrastructure boom faces physical constraints including power limits, supply chain delays, and capital costs.
A Financial Times survey of 94 economists indicates expectations for an AI-driven global economic surge extending beyond the US and China.
Anthropic is engaging legal experts to establish legal frameworks and liability standards for autonomous AI negotiation agents.
Rising financing costs and US tech selloffs threaten earnings across Asia's AI hardware supply chain.
Replit CEO confirms the company ended its partnership with Anthropic, citing factors including cost, following an OpenAI deal in August.
Researcher Nick Cammarata claims AI agents reduce polysemanticity, significantly improving model interpretability and safety alignment.
Researchers propose TLA-Verus, using LLM agents to generate and solve temporal proof obligations for distributed systems specifications.
Researchers propose test-time adaptation methods for AI text detectors to handle post-deployment distribution shifts and new LLMs.
Researchers introduce NeoHorse-1, an agent-native model family exploring recursive self-improvement via agentic post-training and routing.
Fiorillo v0.5, a 4-billion-parameter open model fine-tuned for evidence inference on randomized trials, has been released.