According to @AndrewYNg, the AI Dev 25 x NYC agenda will feature developers from Google, AWS, Vercel, Groq, Mistral AI, and SAP sharing lessons from building production AI systems (source: @AndrewYNg on X). Key topics include agentic architecture trade-offs, autonomous planning for edge cases, and when orchestration frameworks help versus when they accumulate errors (source: @AndrewYNg on X). The program highlights context engineering limits of retrieval for complex reasoning, how knowledge graphs connect information that vector search misses, and building memory systems that preserve relationships (source: @AndrewYNg on X). Infrastructure sessions address scaling bottlenecks across hardware, models, and applications, semantic caching strategies that cut costs and latency, and how faster inference enables better orchestration (source: @AndrewYNg on X; ai-dev.deeplearning.ai). Production-readiness and tooling tracks cover systematic agent testing, translating AI governance into engineering practice, MCP implementations, context-rich code review systems, and adaptable demos (source: @AndrewYNg on X). For traders tracking AI infrastructure equities and AI-crypto narratives, the agenda emphasizes latency, cost optimization, and orchestration efficiency as current enterprise priorities, which can guide sentiment monitoring and thematic positioning (source: @AndrewYNg on X).

According to Hyperbolic (@hyperbolic_labs), LLoCO on NVIDIA H100 delivered up to 7.62x faster inference on 128k-token sequences and 11.52x higher throughput during finetuning, and enabled processing of 128k tokens on a single H100 (source: Hyperbolic on X, Aug 21, 2025). For trading context, these stated gains are concrete performance datapoints for assessing throughput per H100 in long-context LLM workloads and may inform evaluation of AI compute efficiency tied to H100 deployments (source: Hyperbolic on X, Aug 21, 2025).

According to Hyperbolic, LLoCO delivered up to 7.62x faster inference on 128k-token sequences on Nvidia H100 GPUs, based on their reported results; source: Hyperbolic @hyperbolic_labs, Aug 20, 2025. According to Hyperbolic, LLoCO achieved 11.52x higher throughput during finetuning on H100; source: Hyperbolic @hyperbolic_labs, Aug 20, 2025. According to Hyperbolic, LLoCO enabled processing of 128k tokens on a single H100; source: Hyperbolic @hyperbolic_labs, Aug 20, 2025.