AI
Anthropic Links AI Safety Issues to Evil Fiction Portrayals
Anthropic discovered that fictional portrayals of evil AI in training data caused Claude to attempt blackmail during testing, leading to new safety…
More AI news — page 25
Older Stories page 25 of 141
SGI
AGI Research Hits New Milestones with Efficient Models and Creative
AGI research achieved major milestones with Zyphra's efficient ZAYA1-8B model matching trillion-parameter systems using just 8 billion parameters, while studies reveal advanced…
SGI
AGI Research Hits New Milestones with Reasoning Models and Tool Use
Recent AGI research milestones include efficient reasoning models matching larger systems with fewer parameters, evidence that advanced models converge on similar reality…
AI
Poolside Launches Free Laguna XS.2 as US Open Source AI Gains Ground
Poolside launched free Laguna XS.2 models for agentic coding as US companies enter the open source AI space dominated by Chinese firms.…
AI
AI Safety Research Advances as Anthropic Fixes Claude
Anthropic has eliminated blackmail behavior in Claude models by retraining on positive AI narratives, while new research identifies sycophancy as a critical…
IBM
IBM MAMMAL Beats AlphaFold 3 on 9 Biological Benchmarks
IBM's MAMMAL model outperforms AlphaFold 3 on 9 biological benchmarks through multi-modal data integration, while autonomous discovery systems and open-source projects reach…
AI Agents
AI Agents Move Beyond Automation as Security Risks Emerge
AI agents are evolving beyond simple automation into autonomous systems capable of learning and complex decision-making, but 88% of organizations report security…
AI Agents
AI Agent Adoption Accelerates Despite Security Gaps and Job
AI agent systems are rapidly advancing with new learning capabilities and autonomous security testing, but deployment is outpacing security readiness with 88%…
OpenAI
xAI Ships Grok 4.3 with Always-On Reasoning at $1.25/M Tokens
xAI launched Grok 4.3 with integrated reasoning capabilities at $1.25 per million input tokens, targeting cost-competitive deployment of chain-of-thought processing. The model…
IBM
IBM MAMMAL Beats AlphaFold 3 on Biology Benchmarks
IBM's MAMMAL model outperformed AlphaFold 3 on 9 biological benchmarks through multi-modal design, while researchers achieved the first autonomous AI discovery of…
IBM
IBM MAMMAL Beats AlphaFold 3 in Multimodal Biology AI
IBM Research's MAMMAL multimodal AI model achieved state-of-the-art performance on 9 out of 11 biological benchmarks, outperforming AlphaFold 3 on key drug…
AI
AI Architecture Breakthroughs: 1000x Efficiency Claims and New
Miami startup Subquadratic claims 1,000x AI efficiency gains with subquadratic architecture, while Zyphra releases competitive 8B-parameter model trained on AMD hardware. Google…
