AI-360

NVIDIA's Q3 fiscal 2025 results to be discussed Nov 20, with CFO Colette Kress releasing commentary prior. Q&A limited to analysts and institutional investors.

"ChatGPT-4 scored 92 in clinical reasoning vs physicians' 74-76. AI-assisted doctors completed diagnoses 1+ minute faster but showed no accuracy gains"

"Agentforce enables autonomous AI agents triggered by data/rules, built via no-code tools, integrated with Data Cloud/MuleSoft/Slack for 24/7 automation"

"Decagon combines GPT-3.5 for query rewriting in RAG workflows with GPT-4 for complex decisions, achieving 91% automation in customer support"

"Claude 3.5 Sonnet scored 93.7% on HumanEval, tops SWE-bench Verified, offers real-time debugging and test generation via Amazon Bedrock inference"

"Enterprise RAs combine NVIDIA-Certified servers, Spectrum-X AI Ethernet, BlueField-3 DPUs with NeMo and NIM microservices for optimized AI infrastructure"

"Using H100 GPUs & Llama 3, Ntropy achieves 20x model efficiency, while NayaOne's AI sandbox delivers 10x faster fraud detection with 40% cost reduction."

"Voice tech update adds 12 African languages to Search & Gboard, with $1.5M to train Nigerians in AI/data skills via Data Scientists Network Foundation."

"New AI system shows 40% better fraud detection, uses RAPIDS, GNN & XGBoost. Bunq reports 100x faster model training; AmEx makes fraud calls in milliseconds."

"HPT unifies 200,000+ robot trajectories across 52 datasets, processing vision and proprioception inputs equally, improving performance by 20% in real tests"

"Studies found fabrications in 80% of public meetings, 50% of 100+ hours of audio, and nearly all 26,000 transcripts analyzed by developers"

"Research spans autonomous surgery systems, smart assistants for medical diagnosis/climate prediction, and AI-art integration for creative interpretation."

"New code sandbox allows Claude to run JavaScript, clean data, create visualisations, and perform real-time analysis on uploaded CSV files."

"NVIDIA plans 10x GPU expansion in India, partnering with tech firms to train 500k developers in sovereign AI, agentic AI, and physical AI systems."

"Quantised Llama 3.2 achieves 56% size reduction using QLoRA and SpinQuant, with 4-bit weights and 8-bit activations for mobile deployment."