Google's FACTS Benchmark Reveals 70% Factuality Ceiling in Enterprise AI Models
New Google benchmark exposes critical limitation in generative AI models' ability to maintain factual accuracy in enterprise applications.
Sources · VentureBeat AI
The news desk
Follow what’s changing.
Recent developments in artificial intelligence, with the original sources one click away.
All news
1521–1540 of 1,902Internal Tensions Emerge Between Meta's AI Team and Veteran Executives
Meta's leading AI researchers are reportedly clashing with longtime company executives, creating an us-versus-them dynamic within the organization.
Sources · NYT Technology
Three New Papers Advance Retrieval-Augmented Generation Research
Researchers publish three arXiv papers addressing RAG applications in activity recognition, hallucination detection, and efficient knowledge retrieval.
Sources · arXiv AI (cs.AI)
Two New arXiv Papers Address Privacy in LLM Alignment and Channel Estimation for 6G Networks
Researchers publish papers on privacy-preserving LLM alignment and applying LLMs to massive MIMO channel estimation for 6G networks.
Sources · arXiv AI (cs.AI)
Accenture Announces Major AI Partnerships with Both Anthropic and OpenAI
Accenture expands AI capabilities through separate strategic partnerships with Anthropic and OpenAI, training tens of thousands of professionals.
Sources · anthropic.com / openai.com
Last Week in AI Podcast Covers DeepSeek 3.2, TPU Ecosystem Shift, and Nested Learning
Episode #227 discusses DeepSeek's new model, NVIDIA partners moving toward Google TPUs, and nested learning concepts in AI architecture.
Sources · Last Week in AI
Mistral Releases Devstral 2 Coding Model; Researchers Explore LLM Agents for Physics Analysis
Mistral launches new coding-focused model while researchers demonstrate LLM automation in particle physics data analysis.
Sources · VentureBeat AI / arXiv AI (cs.AI)
OpenAI Hires Slack CEO Denise Dresser as Chief Revenue Officer
Denise Dresser is leaving Slack to join OpenAI as chief revenue officer, focusing on enterprise revenue strategy and customer success.
Sources · WIRED AI / TechCrunch AI
Researchers Release Frameworks for Evaluating AI Biosecurity and Medical Privacy Risks
New arXiv papers introduce systematic evaluation methods for assessing biological threat risks and security vulnerabilities in frontier AI models.
Sources · arXiv AI (cs.AI)
Scout24 Deploys GPT-5-Powered Assistant to Transform Real Estate Search
Scout24 launches conversational AI assistant using GPT-5 to guide property seekers through personalized search with questions and recommendations.
Sources · OpenAI
OpenAI Faces Trademark Concerns Over Product Naming Choices
WIRED reports OpenAI has repeatedly chosen names for its releases that resemble existing trademarks, including 'cameo' and 'io.'
Sources · WIRED AI
OpenAI Partners with Deutsche Telekom to Deploy AI Across Europe
OpenAI and Deutsche Telekom announce collaboration to bring multilingual AI experiences to millions of European users and deploy ChatGPT Enterprise internally.
Sources · OpenAI
Three New Research Papers Address LLM Limitations in Reasoning, Tool Use, and Serving Efficiency
Recent arXiv papers tackle LLM hallucinations through neuro-symbolic integration, tool-use training data scarcity, and MoE model serving challenges.
Sources · arXiv AI (cs.AI)
Two New Frameworks Target AI Reasoning Challenges in Healthcare and Vision-Language Models
Researchers introduce MCP-AI for healthcare reasoning and TRACE for evaluating vision-language model step-by-step reasoning processes.
Sources · arXiv AI (cs.AI)
Zhipu AI Releases Open-Source GLM-4.6V Vision-Language Models
Chinese AI startup launches GLM-4.6V series optimized for multimodal reasoning and frontend automation with models in two sizes.
Sources · VentureBeat AI
Anthropic Study Reveals AI's Impact on Its Own Engineers' Productivity and Work Patterns
Anthropic surveyed 132 engineers, finding they use Claude for 60% of work tasks and report 50% productivity gains, though concerns about expertise depth emerge.
Sources · anthropic.com
Multiple AI Companies Release New Models: DeepSeek 3.2, Mistral 3, and Runway Gen-4.5
DeepSeek, Mistral, and Runway announced new model releases, with AWS introducing Trainium3 chips, according to Last Week in AI newsletter.
Sources · Last Week in AI
OpenAI Disables ChatGPT App Suggestions Following User Backlash
OpenAI turned off promotional app suggestions in ChatGPT after users complained they resembled advertisements.
Sources · TechCrunch AI
Three New Research Papers Address LLM Challenges in Reasoning, Tool Use, and Efficient Serving
Recent arXiv papers tackle LLM hallucinations through neuro-symbolic integration, improve agent training data, and optimize MoE model serving.
Sources · arXiv AI (cs.AI)
Two New Benchmarks Advance Retrieval-Augmented Generation Research
Researchers introduce RAG-IGBench for evaluating interleaved image-text generation and M4-RAG for multilingual multimodal RAG systems.
Sources · arXiv AI (cs.AI)