Page 7 of 15 · 177 publications
Musubi has introduced PolicyLM-1.7B, a specialized open-weights decision model that evaluates content against plain-English policies in under 50 milliseconds. This approach bypasses costly retraining cycles, bridging the gap between rigid classifiers and flexible LLMs.
OpenAI is rolling out an invisible cryptographic watermarking system for ChatGPT and Codex users within the European Union. Designed to comply with strict EU AI Act transparency mandates, the technique alters token selection without degrading model performance.
Brooklyn-based startup Reflection AI has introduced Beam, a massive 501-billion-parameter open-weight model trained via intensive reinforcement learning. Designed to compete with top-tier Chinese models and Western alternatives, Beam promises high-level reasoning and agentic performance at drastically reduced inference costs.
Google has halted its open-source bug bounty program until early 2027 due to an unsustainable influx of invalid, hallucinated reports driven by automated AI tools. This unprecedented move highlights the growing crisis of generative AI spam undermining software security pipelines.
The U.S. administration has mandated a terminology shift from artificial intelligence to super intelligence, coinciding with a high-profile executive summit and a voluntary safety pact. Industry analysts remain skeptical about the legal weight of these commitments, viewing the initiative primarily as a political rebranding strategy.
Facing mounting regulatory pushback and over a hundred proposed data center moratoriums across the U.S., Amazon Web Services has abandoned the use of nondisclosure agreements with government agencies. In a strategic push for transparency, leadership is attempting to dispel common myths surrounding grid strain, water consumption, and community impact.
A veteran OpenAI safety team member has stepped down, publishing a critical essay that argues the artificial intelligence industry's rapid deployment culture is fundamentally broken. The departure underscores rising internal anxieties regarding how frontier labs govern increasingly autonomous and capable machine learning models.
Meta is taking its consumer-focused AI agent, Muse, beyond software with the launch of Muse Gadgets. This new open-source initiative provides developers with firmware, a Linux SDK, and hardware blueprints to build custom agentic physical devices.
Sean Parker and CEO Prem Akkaraju are steering Stability AI away from its turbulent past toward the music industry. Backed by major record labels, the company is launching advanced audio generation and editing models.
Recent investigative reports reveal that Elon Musk's Grok chatbot provided strategic scenario analysis to high-ranking officials prior to major military operations. This convergence of generative AI and statecraft highlights a rapidly evolving paradigm where LLMs directly inform geopolitical decisions.
Alphabet has introduced Gemini 4 Argon, an advanced foundation model engineered to handle long-horizon tasks, software engineering, and autonomous cybersecurity defense. Initially restricted to select security partners, the model leverages deep reasoning to identify and patch complex system vulnerabilities.