{
  "version": "https://jsonfeed.org/version/1.1",
  "title": "ML & AI in Action",
  "home_page_url": "https://ajing.github.io/",
  "feed_url": "https://ajing.github.io/feed.json",
  "description": "Explore the Practical Side of ML & AI: Real-World Applications and Impact.",
  "authors": [
    {
      "name": "Jing Lu",
      "url": "https://ajing.github.io/"
    }
  ],
  "language": "en",
  "items": [
    {
      "id": "https://ajing.github.io/posts/2026-05-28-agent-eval-difficulty-trajectory-constraints/",
      "url": "https://ajing.github.io/posts/2026-05-28-agent-eval-difficulty-trajectory-constraints/",
      "title": "How to Arbitrarily Increase the Difficulty of Agent Evaluation Sets",
      "summary": "A practical framework for making agent benchmarks harder in a controlled way: treat difficulty as trajectory-graph complexity, not prompt wording. Covers deterministic scoring, capability facets, harness effects, and systematic data generation.",
      "content_text": "A practical framework for making agent benchmarks harder in a controlled way: treat difficulty as trajectory-graph complexity, not prompt wording. Covers deterministic scoring, capability facets, harness effects, and systematic data generation.",
      "date_published": "2026-05-28T08:00:00.000Z",
      "authors": [
        {
          "name": "Jing Lu"
        }
      ],
      "tags": [
        "AI",
        "LLM",
        "ML Engineering",
        "Agents",
        "Evaluation"
      ]
    },
    {
      "id": "https://ajing.github.io/posts/2026-04-13-mercor-breach-what-leaked-about-ai-training/",
      "url": "https://ajing.github.io/posts/2026-04-13-mercor-breach-what-leaked-about-ai-training/",
      "title": "The Mercor Breach: What 4TB of Stolen Data Reveals About How Frontier AI Labs Actually Train Models",
      "summary": "A $10B AI data vendor was breached, exposing 84 Airtable workspaces of training data for OpenAI, Anthropic, Apple, Amazon, and Meta. This post analyzes what the public reporting reveals about each lab's evaluation methodology — rubric design, RLHF pipelines, and quality control — and what it means for the industry.",
      "content_text": "A $10B AI data vendor was breached, exposing 84 Airtable workspaces of training data for OpenAI, Anthropic, Apple, Amazon, and Meta. This post analyzes what the public reporting reveals about each lab's evaluation methodology — rubric design, RLHF pipelines, and quality control — and what it means for the industry.",
      "date_published": "2026-04-13T00:00:00.000Z",
      "authors": [
        {
          "name": "Jing Lu"
        }
      ],
      "tags": [
        "AI",
        "RLHF",
        "ML Engineering",
        "LLM",
        "Security",
        "Post-Training"
      ]
    },
    {
      "id": "https://ajing.github.io/posts/2026-03-11-improving-llm-i18n-tool-use-agency/",
      "url": "https://ajing.github.io/posts/2026-03-11-improving-llm-i18n-tool-use-agency/",
      "title": "Improving LLM Internationalization: Bridging the Gap in Tool Use and Agency",
      "summary": "LLMs achieve 57% tool-calling accuracy in English but only 34% across 52 languages — and 6.8% for the worst. This post covers the full playbook for closing the multilingual gap: training-time techniques, agentic architecture patterns, failure mode analysis, and RL-based approaches for i18n.",
      "content_text": "LLMs achieve 57% tool-calling accuracy in English but only 34% across 52 languages — and 6.8% for the worst. This post covers the full playbook for closing the multilingual gap: training-time techniques, agentic architecture patterns, failure mode analysis, and RL-based approaches for i18n.",
      "date_published": "2026-03-11T00:00:00.000Z",
      "authors": [
        {
          "name": "Jing Lu"
        }
      ],
      "tags": [
        "AI",
        "LLM",
        "ML Engineering",
        "Agents",
        "i18n"
      ]
    },
    {
      "id": "https://ajing.github.io/posts/2026-03-08-unverifiable-rewards-rl-frontier/",
      "url": "https://ajing.github.io/posts/2026-03-08-unverifiable-rewards-rl-frontier/",
      "title": "The Unverifiable Reward Problem: The Real Frontier of RL for LLMs",
      "summary": "Deep research on tasks with unverifiable rewards in RL — the key bottleneck for scaling RL beyond math and code. Covers JEPO, NRT, RLNVR, self-play methods, GenRM, Constitutional AI, reward hacking mitigation, and more.",
      "content_text": "Deep research on tasks with unverifiable rewards in RL — the key bottleneck for scaling RL beyond math and code. Covers JEPO, NRT, RLNVR, self-play methods, GenRM, Constitutional AI, reward hacking mitigation, and more.",
      "date_published": "2026-03-08T00:00:00.000Z",
      "authors": [
        {
          "name": "Jing Lu"
        }
      ],
      "tags": [
        "AI",
        "RLHF",
        "ML Engineering",
        "LLM",
        "Reinforcement Learning"
      ]
    },
    {
      "id": "https://ajing.github.io/posts/2026-03-05-instruction-following-post-training-data/",
      "url": "https://ajing.github.io/posts/2026-03-05-instruction-following-post-training-data/",
      "title": "Instruction Following: What Models Get Wrong and How to Fix It with Better Post-Training Data",
      "summary": "LLMs can write poetry and solve math, but ask them to 'respond in exactly 3 bullet points using only lowercase' and they stumble. This post dissects the taxonomy of instruction-following failures and provides a practical playbook for building post-training data that actually fixes them.",
      "content_text": "LLMs can write poetry and solve math, but ask them to 'respond in exactly 3 bullet points using only lowercase' and they stumble. This post dissects the taxonomy of instruction-following failures and provides a practical playbook for building post-training data that actually fixes them.",
      "date_published": "2026-03-05T00:00:00.000Z",
      "authors": [
        {
          "name": "Jing Lu"
        }
      ],
      "tags": [
        "AI",
        "LLM",
        "ML Engineering",
        "Post-Training"
      ]
    },
    {
      "id": "https://ajing.github.io/posts/2026-03-01-experience-augmented-icl-complement-to-rl-post-training/",
      "url": "https://ajing.github.io/posts/2026-03-01-experience-augmented-icl-complement-to-rl-post-training/",
      "title": "Experience-Augmented In-Context Learning: A Training-Free Complement to RL Post-Training",
      "summary": "RL post-training makes models smarter, but it can't cover the infinite long tail of real-world cases. Experience-augmented ICL retrieves successful reasoning traces at inference time, letting agents learn continuously from real usage — no retraining required.",
      "content_text": "RL post-training makes models smarter, but it can't cover the infinite long tail of real-world cases. Experience-augmented ICL retrieves successful reasoning traces at inference time, letting agents learn continuously from real usage — no retraining required.",
      "date_published": "2026-03-01T00:00:00.000Z",
      "authors": [
        {
          "name": "Jing Lu"
        }
      ],
      "tags": [
        "AI",
        "LLM",
        "ML Engineering",
        "Agents",
        "RAG"
      ]
    },
    {
      "id": "https://ajing.github.io/posts/2026-01-10-tool-selection-optimization-llm-agents-at-scale/",
      "url": "https://ajing.github.io/posts/2026-01-10-tool-selection-optimization-llm-agents-at-scale/",
      "title": "Tool Selection Optimization for LLM Agents at Scale",
      "summary": "A deep technical dive into tool selection—retrieval strategies, context optimization, learned selection, and the engineering trade-offs that matter when scaling to hundreds of tools.",
      "content_text": "A deep technical dive into tool selection—retrieval strategies, context optimization, learned selection, and the engineering trade-offs that matter when scaling to hundreds of tools.",
      "date_published": "2026-01-10T00:00:00.000Z",
      "authors": [
        {
          "name": "Jing Lu"
        }
      ],
      "tags": [
        "AI",
        "LLM",
        "ML Engineering",
        "Agents"
      ]
    },
    {
      "id": "https://ajing.github.io/posts/2026-01-03-generative-engine-optimization-geo/",
      "url": "https://ajing.github.io/posts/2026-01-03-generative-engine-optimization-geo/",
      "title": "Generative Engine Optimization (GEO): How to Get Your Product Cited by AI",
      "summary": "A comprehensive guide to Generative Engine Optimization—making your content retrievable, citable, and recommendable by large language models.",
      "content_text": "A comprehensive guide to Generative Engine Optimization—making your content retrievable, citable, and recommendable by large language models.",
      "date_published": "2026-01-03T00:00:00.000Z",
      "authors": [
        {
          "name": "Jing Lu"
        }
      ],
      "tags": [
        "AI",
        "Marketing",
        "GEO",
        "SEO",
        "LLM"
      ]
    },
    {
      "id": "https://ajing.github.io/posts/2026-01-02-ads-formats-in-llm-products/",
      "url": "https://ajing.github.io/posts/2026-01-02-ads-formats-in-llm-products/",
      "title": "Ad Formats in LLM Products: What's Live vs. What's Research",
      "summary": "A survey of advertising formats in LLM products—separating what's deployed in production from what remains in research.",
      "content_text": "A survey of advertising formats in LLM products—separating what's deployed in production from what remains in research.",
      "date_published": "2026-01-02T00:00:00.000Z",
      "authors": [
        {
          "name": "Jing Lu"
        }
      ],
      "tags": [
        "AI",
        "LLM",
        "Advertising",
        "Product"
      ]
    },
    {
      "id": "https://ajing.github.io/posts/2026-01-02-ads-in-llm-chatbot/",
      "url": "https://ajing.github.io/posts/2026-01-02-ads-in-llm-chatbot/",
      "title": "Adding Ads in LLM/Chatbot: Character Training for Monetization",
      "summary": "Exploring how to integrate ads in LLMs through character training—making recommendations genuinely helpful rather than annoyingly promotional.",
      "content_text": "Exploring how to integrate ads in LLMs through character training—making recommendations genuinely helpful rather than annoyingly promotional.",
      "date_published": "2026-01-02T00:00:00.000Z",
      "authors": [
        {
          "name": "Jing Lu"
        }
      ],
      "tags": [
        "AI",
        "LLM",
        "Advertising",
        "RLHF"
      ]
    },
    {
      "id": "https://ajing.github.io/posts/2026-01-01-generative-ui-doesnt-move-the-needle-steering-does/",
      "url": "https://ajing.github.io/posts/2026-01-01-generative-ui-doesnt-move-the-needle-steering-does/",
      "title": "Generative UI Doesn't Move the Needle—Steering Does",
      "summary": "After shipping multiple generative UI features, I've concluded that the sophistication of AI-generated interfaces often doesn't translate to user benefit—but steering does.",
      "content_text": "After shipping multiple generative UI features, I've concluded that the sophistication of AI-generated interfaces often doesn't translate to user benefit—but steering does.",
      "date_published": "2026-01-01T00:00:00.000Z",
      "authors": [
        {
          "name": "Jing Lu"
        }
      ],
      "tags": [
        "AI",
        "UX",
        "LLM",
        "Product"
      ]
    },
    {
      "id": "https://ajing.github.io/posts/2025-12-31-rlhf-engineering-implementation/",
      "url": "https://ajing.github.io/posts/2025-12-31-rlhf-engineering-implementation/",
      "title": "RLHF from an Engineering Perspective: PPO, GRPO, DPO, and Tool-Use Implementation",
      "summary": "A practical engineering guide to RLHF implementation—covering PPO, GRPO, DPO, and tool-use training with code snippets and debugging tips.",
      "content_text": "A practical engineering guide to RLHF implementation—covering PPO, GRPO, DPO, and tool-use training with code snippets and debugging tips.",
      "date_published": "2025-12-31T00:00:00.000Z",
      "authors": [
        {
          "name": "Jing Lu"
        }
      ],
      "tags": [
        "AI",
        "RLHF",
        "ML Engineering",
        "LLM"
      ]
    },
    {
      "id": "https://ajing.github.io/posts/2025-12-31-rlhf-ppo-dpo-grpo-notes/",
      "url": "https://ajing.github.io/posts/2025-12-31-rlhf-ppo-dpo-grpo-notes/",
      "title": "Post-Training Is Not 'One Algorithm': Objective Functions and Implementation Essentials for PPO / DPO / GRPO",
      "summary": "Reading notes on RLHF covering PPO, DPO, and GRPO—understanding post-training as an engineering pipeline rather than a single algorithm.",
      "content_text": "Reading notes on RLHF covering PPO, DPO, and GRPO—understanding post-training as an engineering pipeline rather than a single algorithm.",
      "date_published": "2025-12-31T00:00:00.000Z",
      "authors": [
        {
          "name": "Jing Lu"
        }
      ],
      "tags": [
        "AI",
        "RLHF",
        "ML Engineering",
        "LLM"
      ]
    },
    {
      "id": "https://ajing.github.io/posts/2025-09-07-ae-vae-training-learnings/",
      "url": "https://ajing.github.io/posts/2025-09-07-ae-vae-training-learnings/",
      "title": "What Worked (and What Didn't) When Training AEs and VAEs for Embedding Compression",
      "summary": "Practical lessons from training autoencoders and VAEs for embedding compression—covering dimension choice, KL scheduling, contrastive signals, and evaluation metrics.",
      "content_text": "Practical lessons from training autoencoders and VAEs for embedding compression—covering dimension choice, KL scheduling, contrastive signals, and evaluation metrics.",
      "date_published": "2025-09-07T00:00:00.000Z",
      "authors": [
        {
          "name": "Jing Lu"
        }
      ],
      "tags": [
        "ML",
        "Autoencoder",
        "VAE",
        "Embeddings"
      ]
    },
    {
      "id": "https://ajing.github.io/posts/2025-09-07-user-interest-modeling-with-transformer-architectures/",
      "url": "https://ajing.github.io/posts/2025-09-07-user-interest-modeling-with-transformer-architectures/",
      "title": "User Interest Modeling with Transformer Architectures",
      "summary": "Exploring position embeddings, architecture choices, and training techniques for Transformer-based recommender systems.",
      "content_text": "Exploring position embeddings, architecture choices, and training techniques for Transformer-based recommender systems.",
      "date_published": "2025-09-07T00:00:00.000Z",
      "authors": [
        {
          "name": "Jing Lu"
        }
      ],
      "tags": [
        "ML",
        "Transformer",
        "RecSys",
        "Embeddings"
      ]
    },
    {
      "id": "https://ajing.github.io/posts/2025-05-03-ui-representation-action-execution/",
      "url": "https://ajing.github.io/posts/2025-05-03-ui-representation-action-execution/",
      "title": "UI Representation and Action Execution for Generative UI",
      "summary": "Exploring structured UI representation using JSON Schema, and how to implement action handlers for generative UI systems.",
      "content_text": "Exploring structured UI representation using JSON Schema, and how to implement action handlers for generative UI systems.",
      "date_published": "2025-05-03T00:00:00.000Z",
      "authors": [
        {
          "name": "Jing Lu"
        }
      ],
      "tags": [
        "AI",
        "UI",
        "LLM",
        "Generative"
      ]
    },
    {
      "id": "https://ajing.github.io/posts/2025-04-01-generative-ui/",
      "url": "https://ajing.github.io/posts/2025-04-01-generative-ui/",
      "title": "A Path Towards Generative UI",
      "summary": "Exploring how LLMs can dynamically generate user interfaces that adapt in real time to user needs—the vision behind generative UI.",
      "content_text": "Exploring how LLMs can dynamically generate user interfaces that adapt in real time to user needs—the vision behind generative UI.",
      "date_published": "2025-04-01T00:00:00.000Z",
      "authors": [
        {
          "name": "Jing Lu"
        }
      ],
      "tags": [
        "AI",
        "UI",
        "LLM",
        "Flutter"
      ]
    },
    {
      "id": "https://ajing.github.io/posts/2023-08-20-how-to-make-llm-serving-faster/",
      "url": "https://ajing.github.io/posts/2023-08-20-how-to-make-llm-serving-faster/",
      "title": "How to Make LLM Inference Faster",
      "summary": "An overview of LLM inference optimization techniques including KV cache, FlashAttention, and memory management strategies.",
      "content_text": "An overview of LLM inference optimization techniques including KV cache, FlashAttention, and memory management strategies.",
      "date_published": "2023-08-20T00:00:00.000Z",
      "authors": [
        {
          "name": "Jing Lu"
        }
      ],
      "tags": [
        "LLM",
        "Inference",
        "Performance",
        "ML"
      ]
    }
  ]
}