Tag: #ai

← All posts

• What DeepSeek V4 Flash Actually Changes — A Cost-per-Capability Question, Not a Leaderboard One

  On July 31, 2026, DeepSeek moved its V4-Flash API into public beta. The architecture is identical to April's preview — 284B total parameters, 13B active MoE, 1M context — and the only thing that changed is post-training.

  2026-07-31 · 11 min read #ai#llm#deepseek#inference-cost#moe

• Building an Internal Knowledge Base on an LLM — Permission-Aware Retrieval, Freshness, and the Eval Set You Need Before Launch

  The account Cerebras published on July 15, 2026 of how they built their internal knowledge base exposes the internals of a system fielding more than 15,000 queries a day from humans, automation, and agents alike. But the

  2026-07-31 · 14 min read #ai#rag#enterprise-search#llm#platform-engineering

• How AI Found Two Years' Worth of Chrome Bugs in a Single Month — Reading the Number 1,072 Correctly

  According to an announcement Google made on July 30, 2026, Chrome 149 and 150 — the two versions released in June alone — fixed 1,072 security bugs, more than the 1,036 fixed across the previous two years and 23 mileston

  2026-07-31 · 12 min read #ai#security#chrome#fuzzing#vulnerability

• The Economics of Refactoring: When Does It Pay Off — Calculating With Change Frequency

  The Economic Benefit of Refactoring, published on martinfowler.com on July 30, 2026, refactored a 17,000-line module in 15 steps and measured input tokens for the same repeated change request dropping from 159,564 to 27,

  2026-07-31 · 15 min read #refactoring#engineering#technical-debt#ai#metrics

• What It Means to Generate 3D CAD From a Prompt — Mesh vs. B-rep, and the Constraint Bottleneck

  A tool called CAID that generates 3D CAD models from a prompt showed up on GeekNews Show GN. The limitations its own creator disclosed summarize the state of this entire field with precision — no automatic dimension veri

  2026-07-31 · 12 min read #ai#cad#llm#manufacturing#geometry

• Eleven Days of Bun's Zig-to-Rust Rewrite — What Actually Transfers from a Large AI Migration

  Follows Bun's 11-day move of 535,000 lines of Zig to Rust through the primary sources. From May 3 to May 14, 2026, it took up to 64 Claude instances, about 50 workflows, 6,502 commits, and roughly 165,000 dollars at API

  2026-07-31 · 13 min read #ai#bun#rust#migration#testing

• RAG · Fine-Tuning · Long-Context — Which One for My Problem: What the Papers Actually Measured, and What No One Measured ♪ Listenable

  It's the most common question in LLM architecture, yet most answers are decision trees with no sources. This post answers only with what has been measured. That fine-tuning fails at injecting new knowledge has been measu

  2026-07-17 · 23 min read #rag#llm#fine-tuning#long-context#ai

• How to Actually Cut Your LLM API Bill — Why "90% Off From Caching" Is 25% on the Invoice ♪ Listenable

  A prompt cache read costs one-tenth of the input price. But that does not take 90% off your invoice. Follow the worked example Anthropic publishes in its own docs and, even with the cache fully warm, the total drops only

  2026-07-17 · 22 min read #llm#cost-optimization#prompt-caching#api#ai

• Browser and Computer-Use Agents: Where They Actually Stand, and What the Benchmarks Really Measure ♪ Listenable

  "A computer-use agent hit 83.5% on OSWorld" and "even the strongest agent finishes only 20.6%" are both facts published in 2026, and both are true. The first is OSWorld 1.0; the second is OSWorld 2.0 from the same team.

  2026-07-17 · 24 min read #ai#computer-use#browser-agents#benchmark#prompt-injection

• How AI Agents Fail in Production — 14 Failure Modes, and Why Retrying Is Not Safe ♪ Listenable

  Putting agents in production hurts in three places. First, failures come from system design, not the model — UC Berkeley's MAST study classified 1642 execution traces into 14 failure modes, and 44.2% of them were system-

  2026-07-17 · 20 min read #ai#agents#observability#reliability#mcp

• AI Coding Agents: What to Use for What — Selection Criteria Verified Only Against the Four Vendors' Official Docs ♪ Listenable

  When you are choosing among Claude Code · Cursor · GitHub Copilot · OpenAI Codex, the question asked most often is "which one is cheapest." Yet the prices these four companies publish cannot answer it — because all four

  2026-07-17 · 24 min read #ai#ai-coding-agent#claude-code#cursor#github-copilot

• Is AI Code Review Actually Worth It — What Measured Evidence Says About Accuracy and False Positives ♪ Listenable

  Marketing copy for AI code review tools is full of numbers like "80% of PRs need no human comment," yet almost none of it publishes precision and false-positive rates alongside. Pull the public measurements together and

  2026-07-17 · 25 min read #ai#code-review#static-analysis#evaluation#software-engineering

• Making Video from a Single Image — Kling·Veo·Sora vs Wan·HunyuanVideo, What to Pick and When ♪ Listenable

  When you are choosing a model to turn a single image plus a prompt into video, what actually decides it is not the polish of the demo reel but three things: price per second, input-image constraints, and licensing. This

  2026-07-17 · 24 min read #ai#video-generation#image-to-video#open-weights#licensing

• How AI Agent Memory Is Actually Built — Four Designs, and What the Benchmarks Really Prove ♪ Listenable

  "Agent memory" is not a single technique but a blur of at least four distinct designs — a file scratchpad, summarization/compaction, vector retrieval, and knowledge graphs. This post pins down what each actually does fro

  2026-07-17 · 21 min read #ai#ai-agent#agent-memory#llm#benchmark

• Simulated Customers Never Walk Away — Where LLM User Simulators Inflate Agent Scores ♪ Listenable

  In τ-bench-style conversational agent benchmarks, the 'user' role is played by another LLM. But this simulator isn't the thing being measured — it's the measuring instrument, and instruments need calibration. Three verif

  2026-07-16 · 21 min read #ai#llm#evaluation#agents#simulation

• PD Disaggregation Doesn't Increase Throughput — What Prefill/Decode Disaggregation Actually Buys You

  PD disaggregation — splitting prefill and decode across separate GPUs — is a design that landed in vLLM, SGLang, and TensorRT-LLM alike in 2026, but wherever you look, all you find are numbers like "2x to 7x." Yet the of

  2026-07-16 · 16 min read #llm#ai#inference#kv-cache#vllm

• MCP Drops Sessions — Reading the Stateless Core in the 2026-07-28 Revision

  The next MCP specification revision, 2026-07-28, is the biggest change since launch. Its core move is stripping state out of the protocol layer — the initialize handshake and Mcp-Session-Id sessions disappear, and every

  2026-07-16 · 16 min read #mcp#ai#protocol#agents#integration

• One Issue, the Whole Supply Chain — How an Agent Inside CI Broke, and What the Defenses Actually Bought ♪ Listenable

  The Claude Code GitHub Actions vulnerability that GMO Flatt Security researcher RyotaK disclosed on June 1, 2026 traces, end to end, how an agent dropped into a CI pipeline can become the channel that hands over an entir

  2026-07-16 · 21 min read #security#ai#prompt-injection#supply-chain#ci-cd

• From Documents to a Knowledge Graph: An Honest Pipeline

  'Extract a knowledge graph from your documents' looks like a single LLM call in a demo. But turning a customer's documents into a graph you can actually query is a six-stage pipeline, and most of the cost and pain lives

  2026-07-15 · 10 min read #knowledge-graph#ai#llm#data-engineering

• Graph RAG, Explained: What It Is and When It Earns Its Cost

  The standard RAG recipe — chunk, embed, retrieve top-k — works when the answer sits inside a single chunk, but it stalls structurally on multi-hop questions and on global sensemaking questions that span the whole corpus

  2026-07-15 · 8 min read #rag#graph-rag#knowledge-graph#ai#llm