Post-training Nemotron 3.5 Super VL for scientific accuracy with RL
NVIDIA released a recipe to customize Nemotron 3.5 Super VL using reinforcement learning and distillation, boosting drug-metabolism prediction accuracy by 29%.
RAG, evaluations, inference, fine-tuning and shipping LLM features to production.
NVIDIA released a recipe to customize Nemotron 3.5 Super VL using reinforcement learning and distillation, boosting drug-metabolism prediction accuracy by 29%.
A new analysis reveals that AI-generated formal proofs in Lean require rigorous verification steps beyond compilation to ensure they actually match the intended mathematical statements.
New decision models like Jev offer millisecond responses at a fraction of LLM costs by choosing from predefined options, reducing hallucinations but limiting complex reasoning.
AWS integrates AgentCore memory with Managed Knowledge Bases to enable personalized, context-aware RAG while enforcing strict document access controls.
Inference engineers can manage latency-cost tradeoffs or push performance boundaries using techniques like disaggregation and speculative decoding.
Independent tests show Claude Opus 5.5 achieves identical scores to the newer Fable 5.2 on coding tasks while costing significantly less and running faster.
A technical deep dive into reducing training time and memory usage when fine-tuning Qwen3.6-27B on 70K-token examples using FSDP, LoRA, and shared instruction sets.
A developer recreated the DeepSeek V3 architecture in pure JAX, training a 36M parameter model on a MacBook Air to demonstrate key mechanisms like MLA and MoE.
Live web search helps agents find current data, but ranking logic determines whether internal or external sources win. Hybrid retrieval solves conflicts between stale docs and fresh changelogs.
NVIDIA exposes cuDLA to Python via cuda-bindings 13.3.0, enabling hybrid GPU-DLA workflows on Jetson AGX Orin and RTX Spark laptops.
A technical guide details how to serve the 753-billion-parameter GLM-5.3 model across two NVIDIA DGX Stations using tensor parallelism, RDMA, and vLLM.
Amazon Bedrock AgentCore V2 uses microVM snapshots to reduce session startup latency. Engineers can optimize performance by shifting initialization work to the build phase.
Diogo Almeida and engineers from Glean, UiPath, and Workday demonstrate how fast, non-generative models like Jev enable reliable software automation at the AI Engineer Code Summit.
New analysis from Apple ML defines the maximum reward gain possible under KL constraints, showing best-of-N sampling nears this limit while PPO remains suboptimal.
Google released the Gemini Decisions API on November 12, 2026, enabling low-latency, low-cost evaluation of up to 1,000 closed-set options without generating text.