My Local LLM Scored 6/6. It Was Wrong Every Time
Developer Mark Hall found his local LLM scored 6/6 on a test but every answer was factually wrong, exposing the risk of over-relying on AI without human verification.
#ai
AI
2101 stories tagged AI · page 25 of 88
Kernel Forge: AI System Automates CUDA Kernel Optimization for Faster Machine Learning
Researchers introduced Kernel Forge, an AI system that uses large language models to automatically generate and optimize CUDA GPU kernels, reducing the need for expert programmers and cutting development time for machine learning applications.
GPT-5.6: OpenAI's latest model balances intelligence and efficiency
OpenAI's GPT-5.6 improves AI efficiency across models, inference, and workflows, delivering more useful intelligence per dollar. It offers significant speed and cost improvements over previous versions.
Google DeepMind Expands Gemini API with 3.6 Flash and Custom Hooks
Google DeepMind has introduced new features to its Gemini API, including 3.6 Flash for faster responses and custom hooks for developers. These updates aim to make AI agents more reliable and production-ready.
Cyera Acquires Oasis Security for $1B to Protect AI Agents
Cyera is buying Oasis Security for $1 billion to enhance its ability to secure AI agents. This is Cyera's third acquisition this year, highlighting the growing importance of AI security.
Artists Win Legal Battles Against AI Companies Using Their Work Without Permission
Artists are successfully suing AI companies like Google, Meta, and Anthropic for using their work without consent. Recent court victories show that AI training data practices are under legal scrutiny, setting precedents for future cases.
LoRA Fails to Match Full Fine-Tuning for Multi-Step Procedural Tasks, New Study Finds
A new study from ArXiv cs.AI finds that LoRA, a popular parameter-efficient fine-tuning method, fails to internalize multi-step procedures with conditional branching, uniformly underperforming full fine-tuning across all tested ranks (16–128) on a travel booking task.
Spectral Flow Certificates (SFCs): A Fast Pre-Training Check for Graph Neural Network Suitability
Researchers have developed Spectral Flow Certificates (SFCs), a method to quickly check if a graph's structure will allow information to travel far enough for a Graph Neural Network (GNN) to solve long-range tasks. This could save time and resources by identifying problematic graphs before training begins.
Perplexity’s Personal Computer turns Windows PCs into AI agents
Perplexity has launched its Personal Computer tool for Windows, turning PCs into AI agents that can access local files and apps. This expands access to a powerful AI assistant that was previously only available on Macs.
Attention-Guided Layer Selection Improves LLM Factuality on TruthfulQA
Researchers introduced three new attention-guided strategies—Attention-JSD, Attention-Entropy-Max, and Attention-Entropy-Min—to improve the factual accuracy of large language models by leveraging self-attention mechanisms for better layer selection during contrastive decoding.
Microsoft Unveils AI Security Tools It Says Outperform Competitors and Cost Less
Microsoft launched new AI-powered security tools that it claims are more effective and cheaper than competing platforms from Palo Alto Networks and CrowdStrike. The tools include advanced threat detection, automated response, and AI-driven analytics.
Lost in Context: AI Models Doubt Themselves Before They Should
A new arXiv study reveals that frontier reasoning models fail solvable problems due to 'context anxiety'—premature self-doubt caused by an inability to estimate required tokens. This leads to significant efficiency losses and shorter, less accurate responses.
LiquidAI's LFM2.5-Encoders Speed Up Long-Context AI Inference on Standard CPUs
LiquidAI released LFM2.5-Encoders, an open-source tool that accelerates long-context AI model inference on standard CPUs, reducing the need for expensive GPUs.
jcode: Open-Source Coding Agent Boots 245x Faster Than Claude Code
A developer open-sourced jcode, a coding agent harness that boots 245 times faster than Claude Code, rendering its first frame in 14 milliseconds and using only 27.8 MB of RAM per active session.
Hugging Face models enable nonconsensual deepfake undressing, report finds
A new report by AI Forensics reveals that seven of the top nine image editing models on Hugging Face can generate nonconsensual deepfake images that undress women and children. The platform's content moderation is failing to prevent this misuse.
How Hackers Breached Frontier Lab's AI Agents in the July 2026 Incident
A detailed technical timeline from Hugging Face reveals how hackers compromised Frontier Lab's AI agents in July 2026, exploiting authentication vulnerabilities to access sensitive data before the breach was contained.
OpenAI Report: AI Coding Agents Analyze 10,000 Genomes in Hours, Accelerating Scientific Discovery
A new OpenAI field report details how scientists use AI coding agents to modernize scientific computing, accelerating software development and discovery in genomics and beyond.
GAND: A New Benchmark to Reduce Gender Bias in Machine Translation
Researchers introduced GAND, a gender-ambiguous natural data benchmarking resource for machine translation, designed to help AI systems translate sentences without clear gender cues more accurately and inclusively.
FrED: New Black-Box Method Traces AI Training Data Without Accessing Model Weights
Researchers propose FrED, a probabilistic framework that traces AI training data origins using knowledge graphs without needing access to model weights or internal architecture. This black-box approach could improve transparency and accountability in generative AI.
AutoThinkSQL: AI That Learns When to Skip Reasoning for Faster Database Queries
Researchers introduced AutoThinkSQL, a framework that trains AI models to decide when to use complex reasoning and when to skip it for Text-to-SQL tasks, reducing inference time on simple queries without sacrificing accuracy on complex ones.
AI Models Hide Reasoning in 'Filler' Tokens, Study Finds
A new arXiv study reveals that frontier AI models use semantically irrelevant filler tokens to improve reasoning accuracy by up to 13 percentage points, raising transparency concerns for AI safety.
AI Finds Hidden Errors in Medical Records, But Struggles at Scale
A new study on ArXiv shows that a two-stage Gemini AI pipeline can detect internal inconsistencies in 3,000 MIMIC-IV-Note discharge summaries, but recurring failure modes limit its reliability for automated clinical use.
Speech + Text AI Predicts Romantic Attraction Better Than Text Alone, Study Finds
A new study combining speech analysis with LLM-based transcript analysis shows improved prediction of romantic attraction in speed dating, but the effect is conditional on context, not universal.
ADAGE: A Language-Agnostic Pipeline for Testing AI Reasoning Without Translation Bias
Researchers introduced ADAGE, a language-agnostic pipeline that creates AI benchmarks in native languages without relying on translations. Validated with benchmarks in Arabic, Amharic, and Japanese, ADAGE combines native-speaker curation with LLM-assisted generation to test culturally grounded analogical reasoning.