AI & Machine Learning– category –
-
“Satisfiable Drift” Problem Emerges in Multi-Turn Reasoning of LLMs, Revealed by Novel DRIFT-Bench Benchmark
AI Accelerator Institute International Overview Researchers developed DRIFT-Bench, a solver-instrumented benchmark with 816 test problems across three constraint domains, to evaluate multi-turn reasoning in open-weight models ranging fro... -
Foundation Models Advance Wireless Communications from PHY Intelligence to Network Autonomy via Multimodal Data Alignment and Agentic RAG Frameworks
arXiv International Overview A preprint explores the application of foundation models in wireless communications, from physical layer intelligence to network autonomy. Contrastive foundation models are discussed for aligning multimodal d... -
Meta Aims for AI Differentiation with Enhanced Health Capabilities, Starting with Muse Spark Integration into Consumer Products
Times of India India Overview Alexandr Wang, Meta's Chief AI Officer, announced that future Meta AI models will differentiate themselves through robust health-related capabilities. The Muse Spark model, released in April, has shown promi... -
2026 LLM Leaderboard Reveals Llama 4 Scout as Fastest at 2600 Tokens/Sec, GPT-5.3 Codex Achieves Lowest Latency at 0.003s
Vellum USA Overview The updated 2026 LLM leaderboard, incorporating data from April 2024 onwards, showcases Llama 4 Scout as the fastest model at 2600 tokens/second, while GPT-5.3 Codex records the lowest latency at 0.003 seconds. Nova M... -
Multimodal Foundation Model SQUALL Integrates Histology with Spatial Molecular Programs, Revolutionizing Cancer Biomarker Profiling and Outcome Prediction
bioRxiv International Overview Researchers have developed SQUALL, a multimodal foundation model that integrates histology with spatial molecular programs to deepen the mechanistic interpretation of histopathological assessments. Pretrain... -
Claude Opus 4.8 Achieves Peak Accuracy of 89.08% in Financial LLM Benchmark, Gemini 3.5 Flash Also Highly Rated
AIMultiple USA Overview In a benchmark evaluating over 40 Large Language Models (LLMs) on complex financial reasoning tasks, Anthropic's Claude Opus 4.8 attained the highest accuracy at 89.08%. Google's Gemini 3.5 Flash also demonstrated... -
Tempus AI Unveils Promising Initial Results from Oncology Multimodal Foundation Model Trained on 2.5M Longitudinal Records, Achieving C-index of 0.802 for OS Prediction
Business Wire USA Overview Tempus AI announced initial results at ASCO 2026 for its multimodal foundation model aimed at oncology insight generation. This transformer-based model, trained on 2.5 million longitudinal records, 450,000 digi... -
A Quarter of Enterprise AI Agent Deployments Fail ROI Targets, Citing Governance and IAM Deficiencies
ITSM.tools 他 UK Overview New 2026 research indicates that 25% of AI agent deployments in large UK enterprises are failing to achieve their expected ROI, primarily due to critical governance gaps. Key challenges include the absence of pu... -
FDA-Cleared AI “Artera AI Breast” Demonstrates Clinical Utility for Breast Cancer Prognosis and Chemotherapy Benefit Prediction at ASCO 2026
The Pathologist USA Overview At ASCO 2026, a study showcased the utility of Artera AI Breast, an FDA-cleared multimodal artificial intelligence model, for predicting prognosis and chemotherapy benefit in postmenopausal women with node-po... -
OpenAI, Anthropic, Microsoft CEOs Call for Stricter Laws Against AI Biothreats to Synthetic DNA/RNA
Financial Express USA Overview CEOs from OpenAI, Anthropic, and Microsoft AI, alongside over 50 other signatories, have pressed the US Congress to implement more stringent regulations on synthetic DNA and RNA to mitigate AI-enabled bioth...
