Friday, May 1, 2026
Latest

Archive — Page 5

67 articles total
Black Forest Labs cuts multimodal AI training costs by 2.8x with Self-Flow

Black Forest Labs cuts multimodal AI training costs by 2.8x with Self-Flow

New technique eliminates dependency on frozen external encoders, removing a fundamental bottleneck in diffusion model scaling.

Microsoft's Phi-4 Matches Larger Models While Using Fraction of Compute

Microsoft's Phi-4 Matches Larger Models While Using Fraction of Compute

A 15 billion parameter multimodal model learns when reasoning is necessary, reducing training data and computational overhead significantly.

New Framework Controls Autonomous AI Agents in Drug Discovery

New Framework Controls Autonomous AI Agents in Drug Discovery

Mozi architecture prevents hallucinations from compounding in high-stakes pharmaceutical pipelines.

Researchers Train LLMs for Medical Dialogue With Tree-Based Reinforcement Learning

Researchers Train LLMs for Medical Dialogue With Tree-Based Reinforcement Learning

New method helps AI ask better diagnostic questions in uncertain clinical scenarios using hierarchical decision processes.

Researchers Enable Autonomous AI Agents to Collaborate Without Predefined Workflows

Researchers Enable Autonomous AI Agents to Collaborate Without Predefined Workflows

New method allows LLM agents to coordinate dynamically, reducing redundant work and cascading failures in multi-agent systems.

Researchers Show LLM Judges Amplify Errors Through Shared Biases

Researchers Show LLM Judges Amplify Errors Through Shared Biases

A new method addresses systematic failures in AI evaluation when multiple language models judge other models.

Alibaba's Qwen3.5-9B Outperforms OpenAI's Larger Model on Standard Hardware

Alibaba's Qwen3.5-9B Outperforms OpenAI's Larger Model on Standard Hardware

Chinese AI researchers release efficient open-source model that rivals 120-billion-parameter competitor while running on consumer laptops.

New Benchmark Tests AI Agents Against Adversarial Trading Scenarios

New Benchmark Tests AI Agents Against Adversarial Trading Scenarios

TraderBench combines expert-verified tasks with live market simulations to measure how robust AI traders perform under real financial pressure.

New Method Removes Sensitive Data From AI Recommendation Systems

New Method Removes Sensitive Data From AI Recommendation Systems

Machine unlearning technique balances privacy protection with model performance in generative recommendation systems.

Researchers Erase Toxic Patterns From LLM Representations

Researchers Erase Toxic Patterns From LLM Representations

New method removes harmful knowledge at the source rather than suppressing surface behaviors.

Researchers Deploy LLM Agents to Automate Financial Crime Detection

Researchers Deploy LLM Agents to Automate Financial Crime Detection

New system uses AI to reduce false positives in anti-money laundering screening, addressing a costly compliance bottleneck.

New Benchmark Tests LLMs on Financial Knowledge and Real-World Tasks

New Benchmark Tests LLMs on Financial Knowledge and Real-World Tasks

FIRE framework evaluates whether AI models can handle both theory and practice in finance.