News

Co-LLM allows general AI to collaborate with expert models, improving factual accuracy. It uses a "switch variable" to defer to specialists at word level, enhancing efficiency.

Stanford HAI held workshop for ASEAN on AI governance. Topics: bias, fairness, trust. ASEAN plans new AI initiatives, including ethics guide and working group.

OpenAI's o1-mini matches larger models in STEM tasks at 80% lower cost. It scores 70% on AIME, reaches 86th percentile on Codeforces, and is 3-5x faster than GPT-4o.

OpenAI's new 'o1' models excel in science, coding, and math. They achieved 83% success on IMO qualifying exams, up from GPT-4's 13%.

Anadol fine-tunes Llama models with 500M+ nature images. "Open source information is like gold for art making," he says, emphasising AI's artistic potential.

UL researchers created AI hardware using molecular-level computing. It processes info via atomic movements, achieving 4.1 TOPS/W efficiency in a 14-bit system.