4 articles
LLMs need moral competence, not just performance, for real-world roles. Challenges include fake reasoning, complex moral factors, and global standards. A roadmap with new evaluations is proposed.
Google released Gemini Pro 3.1, a powerful new AI model praised for significant performance improvements in benchmarks and real professional tasks, amid intense competition.
AI chatbots for UK government info are often too wordy, burying facts. When told to be concise, accuracy drops. They also make unpredictable errors and rarely refuse questions, risking misinformation. Smaller models can match larger ones.
Step 3.5 Flash is a high-performing (81.0 score) open-source MoE model (196B/11B params) excelling in reasoning, coding, agentic tasks, tool-use, and deep research, supporting edge-cloud & local deployment.