Researchers Reveal New Jailbreaking Methods for Large Language Models
Three new studies demonstrate vulnerabilities in AI models through emoji-based attacks, text steganography, and adversarial image perturbations.
Sources · arXiv AI (cs.AI)
The news desk
Follow what’s changing.
Recent developments in artificial intelligence, with the original sources one click away.
All news
1421–1440 of 1,902Three New Studies Examine Large Language Models in Brand Discovery, Decision-Making, and Engineering Applications
Recent arXiv preprints explore LLM cultural biases in brand recommendations, cost-aware multi-agent systems, and numerical equation-solving capabilities.
Sources · arXiv AI (cs.AI)
LG AI Research Releases EXAONE Model Series: From 3.0 to 4.0 with Enhanced Reasoning
LG AI Research introduces EXAONE language models, including the open 7.8B model and new 4.0 version with dual reasoning modes.
Sources · arXiv AI (cs.AI)
New Research Evaluates LLM Capabilities in Code Security, Preference Testing, and Real-World API Usage
Three arXiv papers examine LLM performance in vulnerability detection, output evaluation methods, and function-calling under realistic conditions.
Sources · arXiv AI (cs.AI)
New Research Identifies Critical Challenges in Large Language Model Architecture and Safety
Recent arXiv papers reveal efficiency issues in LLM layer architecture, propose methods for safer multi-turn conversations, and examine safety risks in reasoning models.
Sources · arXiv AI (cs.AI)
Three New arXiv Papers Address LLM Quantization and Specialized Applications
Researchers publish papers on pedestrian behavior inference using LLMs, self-explanation capabilities after quantization, and ultra-low-bit quantization methods.
Sources · arXiv AI (cs.AI)
Two Studies Examine LLMs' Capabilities in IoT Security and Vulnerability Scoring
Research explores whether large language models can detect smart home security threats and automate vulnerability scoring systems.
Sources · arXiv AI (cs.AI)
AI Chatbots Show Mixed Accuracy on Breaking News Events
WIRED reports varying performance among AI chatbots in handling breaking news, with some performing well and others struggling with factual accuracy.
Sources · WIRED AI
Mercor Reaches $10 Billion Valuation Connecting AI Labs with Professional Workers
Three-year-old startup Mercor has achieved a $10 billion valuation by matching AI companies with professionals for data work.
Sources · TechCrunch AI
OpenAI Opens Applications for Grove Cohort 2 Founder Program
OpenAI launches applications for its second Grove cohort, offering founders $50K in API credits, early AI tool access, and mentorship.
Sources · OpenAI
Schools Adopt AI Chatbots Amid Expert Warnings About Educational Impact
Governments are introducing AI chatbots in schools, but experts caution about potential negative effects on teaching and learning quality.
Sources · NYT Technology
TechCrunch Predicts AI Industry Shift Toward Pragmatism in 2026
TechCrunch forecasts AI development will prioritize practical applications, smaller models, and real-world products in 2026.
Sources · TechCrunch AI
Erotic Chatbots Emerge as Defining AI Use Case in 2025, Reports WIRED
Despite productivity promises, erotic chatbots have become AI's dominant narrative in 2025, according to WIRED analysis.
Sources · WIRED AI
OpenAI Focuses on Audio Interfaces as Silicon Valley Shifts Away from Screens
TechCrunch reports OpenAI and other tech companies are prioritizing audio as the next major user interface, moving beyond traditional screens.
Sources · TechCrunch AI
Tech Industry Explores Space-Based Data Centers to Address AI Infrastructure Constraints
Tech leaders consider orbital data centers as potential solution to land and energy constraints facing AI development, according to NYT report.
Sources · NYT Technology
Three New arXiv Papers Examine LLM Capabilities in Question Answering, Recommendation Systems, and Debate Generation
Recent arXiv publications explore LLM limitations in retrieval-augmented QA, collaborative signal understanding, and argumentative debate generation.
Sources · arXiv AI (cs.AI)
Three New Benchmarks Released for Evaluating LLM Capabilities in Commerce, Tool Use, and Building Management
Researchers published three arXiv papers introducing benchmarks for testing LLMs in e-commerce search, tool usage via MCP, and smart building energy systems.
Sources · arXiv AI (cs.AI)
Three New Papers Explore Multi-Step Reasoning in Large Language Models
Recent arXiv preprints examine how LLMs handle complex multi-step tasks in spatial reasoning, retrieval-augmented generation, and radiology.
Sources · arXiv AI (cs.AI)
Three New Research Papers Address LLM Agent Efficiency, Reasoning Speed, and Personalized Search
Researchers publish papers on automating agent creation, accelerating LLM inference, and building personalized search systems using AI agents.
Sources · arXiv AI (cs.AI)
Three New Research Papers Explore Latent Planning and Reasoning in Large Language Models
Researchers propose new approaches for improving LLM reasoning through latent planning, adaptive semantic spaces, and interactive navigation methods.
Sources · arXiv AI (cs.AI)