Subscribe
Subscribe
Archive / page 18 of 49

Latest articles, page 18

  1. Kahneman-Tversky Optimization (KTO): Mathematical Foundations, Prospect Theory, and Binary Signal Alignment in Large Language Models

    Post-training alignment of large language models has long relied on pairwise comparison datasets. Methods such as Reinforcement Learning from Human Feedback (RLHF) and Direct Preference Optimization (DPO) assume access to curated pairs $(x, y_w, y_l)$, where a human or automated judge explicitly marks completion $y_w$ as superior to $y_l$ for a given prompt $x$. In production environments, however, collecting paired preferences is logistically complex, expensive, and artificial. Real-world telem

    Explainers
    10 min read

  2. Keenable Exits Stealth with $26M Seed to Build Web Index for AI Agents

    Keenable Exits Stealth with $26M Seed to Build Web Index for AI Agents Search infrastructure startup Keenable has emerged from stealth with $26 million in seed funding led by Accel, with participation from Conviction Partners and angel investors. The company is developing a web-scale indexing engine tailored specifically for programmatic retrieval by autonomous AI agents and language model workflows rather than human web browsers. Keenable was co-founded by Andrey Styskin, former head of searc

    Search
    2 min read

  3. OpenAI Publishes Benchmark Results for Jalapeño Custom Inference Silicon

    OpenAI Publishes Benchmark Results for Jalapeño Custom Inference Silicon OpenAI has shared its first empirical performance data for Jalapeño, the custom inference processor co-developed with Broadcom. Presented at the Hot Chips conference and published in an engineering report, the benchmark results evaluate the 700-watt accelerator across open-weight frontier models against Nvidia GB200 and GB300 systems using SemiAnalysis's InferenceX test suite. The published metrics focus on serving effici

    Hardware
    3 min read

  4. Multi-Agent Orchestration Frameworks in Production: Comparing LangGraph, AutoGen v0.4, CrewAI, and LlamaIndex Workflows Architecture, State Persistence, Execution Topologies, and Fault Tolerance

    Deploying autonomous multi-agent systems in production requires shifting from simple input-output prompt chains to structured orchestration runtimes capable of handling non-deterministic language model outputs, cyclic reasoning loops, tool-call failures, and long-running state persistence. As enterprise workloads scale, four distinct architectural paradigms have emerged to coordinate agent interactions: the Bulk Synchronous Parallel state graph (LangGraph), the asynchronous Actor Model (Microso

    AI
    8 min read

  5. The Muon Optimizer: Mathematical Foundations, Newton-Schulz Polar Iterations, and Spectral Steepest Descent in LLM Pre-Training

    Transformer pre-training has long relied on AdamW as the default first-order optimizer. While AdamW provides coordinate-wise scale invariance through running estimates of first and second gradient moments, it treats every weight parameter as an isolated scalar. For modern neural networks composed primarily of two-dimensional linear projections, attention transformations, and feed-forward weight matrices, this coordinate-wise factorization neglects the structural transformation properties of matr

    AI
    6 min read

  6. Google Cloud Launches Gemini Enterprise for Legal with Domain Agents and MCP Connectors

    Google Cloud has launched Gemini Enterprise for Legal, a specialized vertical edition of its enterprise AI platform configured specifically for corporate legal departments and law firms. The product debuted in preview on August 25, 2026, alongside a parallel financial services edition, marking Google's initial push into industry-tailored enterprise AI packages. Launch law firms participating in early access include Cleary Gottlieb, Freshfields, Weil, and Williams & Connolly. Architectural Arc

    Google
    2 min read

  7. Perplexity and Nvidia Launch Portable Computer for Local AI Agents with Zero Token Fees

    Perplexity has partnered with Nvidia to launch Portable Computer, an integrated software stack that runs agentic AI workflows locally on consumer and workstation GPUs without incurring per-token API charges. The system packages model weights, an inference server, an agent harness, tool connectors, and an operating system sandbox into a unified application. It is available immediately on Linux for Pro, Max, and Enterprise subscribers, with Windows support scheduled for September. Hardware requir

    Perplexity
    2 min read

  9. Vision-Language Model Serving in Production: Comparing LLaVA-NeXT, Qwen2-VL, InternVL2, and Pixtral Architecture, Dynamic Resolution Patching, Visual Token Pruning, and Serving Latency Economics

    Deploying Vision-Language Models (VLMs) in production introduces systems challenges distinct from text-only Large Language Models. In standard text serving, input length scales linearly with character and token count. In multimodal systems, a single high-resolution image or short video clip can expand into thousands of visual tokens before textual generation begins. This expansion places immense strain on GPU memory bandwidth, inflates Key-Value (KV) cache allocation, and degrades Time-to-First-

    AI
    8 min read

  10. Proximal Policy Optimization: Mathematical Foundations, Clipped Surrogate Objectives, and Policy Drift Control in RLHF

    Reinforcement learning from human feedback (RLHF) transformed autoregressive large language models from raw next-token predictors into instruction-following assistants. At the computational center of the foundational RLHF pipelines introduced in InstructGPT (Ouyang et al., 2022) is Proximal Policy Optimization (PPO), formulated by John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov at OpenAI in 2017. PPO resolved a fundamental instability in policy gradient methods: th

    AI
    11 min read