<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0"><channel><title>AI Explained</title><link>https://ai.miladezzat.com/</link><description>Short, clear explainers on how modern AI works, written by a software engineer for software engineers. Every topic comes with a 2-minute video and a page you can read, save and share.</description><language>en</language>
<item><title>Knowledge cutoff: why AI models don't know what happened yesterday</title><link>https://ai.miladezzat.com/knowledge-cutoff/</link><guid>https://ai.miladezzat.com/knowledge-cutoff/</guid><pubDate>Thu, 19 Nov 2026 12:00:00 GMT</pubDate><description>What an LLM's knowledge cutoff is, why the real cutoff often differs from the reported one, and how developers give a model the present.</description></item>
<item><title>MCP errors: protocol errors vs tool errors (isError)</title><link>https://ai.miladezzat.com/mcp-16-errors/</link><guid>https://ai.miladezzat.com/mcp-16-errors/</guid><pubDate>Thu, 19 Nov 2026 12:00:00 GMT</pubDate><description>In the 2026-07-28 MCP spec, a tool call can fail in two ways. Here's the difference between protocol errors and tool execution errors, and which one the model should see.</description></item>
<item><title>Retries and idempotency for agent tool calls</title><link>https://ai.miladezzat.com/retries-idempotency/</link><guid>https://ai.miladezzat.com/retries-idempotency/</guid><pubDate>Thu, 19 Nov 2026 12:00:00 GMT</pubDate><description>How to let an AI agent retry a failed tool call without booking, charging or emailing twice, using idempotent operations and idempotency keys.</description></item>
<item><title>Self-consistency: ask an LLM many times, then take the vote</title><link>https://ai.miladezzat.com/self-consistency/</link><guid>https://ai.miladezzat.com/self-consistency/</guid><pubDate>Thu, 19 Nov 2026 12:00:00 GMT</pubDate><description>How self-consistency sampling works, why voting over several reasoning paths makes a model more accurate, what it costs, and how to use it in your own app.</description></item>
<item><title>Continuous batching: how LLM servers keep the GPU full</title><link>https://ai.miladezzat.com/continuous-batching/</link><guid>https://ai.miladezzat.com/continuous-batching/</guid><pubDate>Wed, 18 Nov 2026 12:00:00 GMT</pubDate><description>Why static batching wastes GPU time on LLM inference, how iteration-level scheduling fixes it, and how chunked prefill and memory management fit in.</description></item>
<item><title>MCP capabilities: how both sides say what they support</title><link>https://ai.miladezzat.com/mcp-15-capabilities/</link><guid>https://ai.miladezzat.com/mcp-15-capabilities/</guid><pubDate>Wed, 18 Nov 2026 12:00:00 GMT</pubDate><description>In the 2026-07-28 MCP spec, clients declare capabilities on every request and servers declare theirs in server/discover. Here are the fields, the rules and the new extensions map.</description></item>
<item><title>Sycophancy in LLMs: when AI tells you what you want to hear</title><link>https://ai.miladezzat.com/sycophancy/</link><guid>https://ai.miladezzat.com/sycophancy/</guid><pubDate>Wed, 18 Nov 2026 12:00:00 GMT</pubDate><description>Why AI assistants agree with users even when they're wrong, what the research and OpenAI's GPT-4o rollback showed, and how to test your own app for it.</description></item>
<item><title>FlashAttention: faster attention, same answer</title><link>https://ai.miladezzat.com/flash-attention/</link><guid>https://ai.miladezzat.com/flash-attention/</guid><pubDate>Tue, 17 Nov 2026 12:00:00 GMT</pubDate><description>Why attention is limited by memory traffic on GPUs, how FlashAttention's tiling keeps the work in fast on-chip memory, and what FlashAttention-2 and -3 added.</description></item>
<item><title>MCP server/discover: learning what a server can do</title><link>https://ai.miladezzat.com/mcp-14-server-discover/</link><guid>https://ai.miladezzat.com/mcp-14-server-discover/</guid><pubDate>Tue, 17 Nov 2026 12:00:00 GMT</pubDate><description>In the 2026-07-28 MCP spec, server/discover returns a server's supported versions, capabilities and identity in one request. Here's what it returns and when to call it.</description></item>
<item><title>Query rewriting and HyDE: better questions, better retrieval</title><link>https://ai.miladezzat.com/query-rewriting-hyde/</link><guid>https://ai.miladezzat.com/query-rewriting-hyde/</guid><pubDate>Tue, 17 Nov 2026 12:00:00 GMT</pubDate><description>Why user questions often miss the right documents in RAG, how Rewrite-Retrieve-Read and step-back prompting reshape the query, and how HyDE searches with a fake answer.</description></item>
<item><title>Retrieval metrics: recall@k, MRR and nDCG for RAG</title><link>https://ai.miladezzat.com/retrieval-metrics/</link><guid>https://ai.miladezzat.com/retrieval-metrics/</guid><pubDate>Tue, 17 Nov 2026 12:00:00 GMT</pubDate><description>How to measure whether your RAG retrieval finds the right documents, with precision and recall at k, mean reciprocal rank and nDCG, plus a worked example.</description></item>
<item><title>Logprobs: how sure was your LLM about that answer?</title><link>https://ai.miladezzat.com/logprobs/</link><guid>https://ai.miladezzat.com/logprobs/</guid><pubDate>Mon, 16 Nov 2026 12:00:00 GMT</pubDate><description>What token log probabilities are, how to turn them into percentages, how to request them, and four practical ways developers use them.</description></item>
<item><title>Matryoshka embeddings: one vector, many sizes</title><link>https://ai.miladezzat.com/matryoshka-embeddings/</link><guid>https://ai.miladezzat.com/matryoshka-embeddings/</guid><pubDate>Mon, 16 Nov 2026 12:00:00 GMT</pubDate><description>How Matryoshka Representation Learning trains embeddings you can cut short, what OpenAI and Hugging Face measured, and how to truncate them safely.</description></item>
<item><title>MCP _meta: what every MCP request carries in 2026-07-28</title><link>https://ai.miladezzat.com/mcp-13-meta/</link><guid>https://ai.miladezzat.com/mcp-13-meta/</guid><pubDate>Mon, 16 Nov 2026 12:00:00 GMT</pubDate><description>In stateless MCP, every request carries its protocol version, client capabilities and client info in _meta. Here are the keys, the errors and the rules.</description></item>
<item><title>Positional encoding (RoPE): how Transformers know word order</title><link>https://ai.miladezzat.com/rope-positional-encoding/</link><guid>https://ai.miladezzat.com/rope-positional-encoding/</guid><pubDate>Mon, 16 Nov 2026 12:00:00 GMT</pubDate><description>Why Transformers need positional encoding, how the original sine waves worked, how RoPE rotates queries and keys, and how models stretch past their trained length.</description></item>
<item><title>Rate limits and token budgets: why your LLM app gets a 429</title><link>https://ai.miladezzat.com/rate-limits-token-budgets/</link><guid>https://ai.miladezzat.com/rate-limits-token-budgets/</guid><pubDate>Sun, 15 Nov 2026 12:00:00 GMT</pubDate><description>How LLM API rate limits work (requests, input and output tokens, token buckets, spend caps) and how to stay under them with retries, backoff and budgets.</description></item>
<item><title>Slopsquatting: when AI invents a package and attackers claim it</title><link>https://ai.miladezzat.com/slopsquatting/</link><guid>https://ai.miladezzat.com/slopsquatting/</guid><pubDate>Sun, 15 Nov 2026 12:00:00 GMT</pubDate><description>How attackers register the package names AI coding tools hallucinate, what the research found, and the checks that keep you safe.</description></item>
<item><title>Stateless MCP: what changed in the 2026-07-28 spec</title><link>https://ai.miladezzat.com/mcp-12-stateless-mcp/</link><guid>https://ai.miladezzat.com/mcp-12-stateless-mcp/</guid><pubDate>Sun, 15 Nov 2026 12:00:00 GMT</pubDate><description>The 2026-07-28 MCP revision makes every request self-contained. Here's what went away (sessions, callbacks, resumability) and what replaced it.</description></item>
<item><title>Tool search for large tool sets: give agents thousands of tools</title><link>https://ai.miladezzat.com/tool-search/</link><guid>https://ai.miladezzat.com/tool-search/</guid><pubDate>Sun, 15 Nov 2026 12:00:00 GMT</pubDate><description>Why loading every tool definition hurts agents, and how tool search with defer_loading lets a model discover only the tools a task needs.</description></item>
<item><title>Agentic browsers: when your browser clicks and buys for you</title><link>https://ai.miladezzat.com/agentic-browsers/</link><guid>https://ai.miladezzat.com/agentic-browsers/</guid><pubDate>Sat, 14 Nov 2026 12:00:00 GMT</pubDate><description>How AI browsers with agent modes work, why indirect prompt injection makes them risky, and how users and builders can limit the damage.</description></item>
<item><title>Fill-in-the-middle: how code autocomplete sees both sides</title><link>https://ai.miladezzat.com/fill-in-the-middle/</link><guid>https://ai.miladezzat.com/fill-in-the-middle/</guid><pubDate>Sat, 14 Nov 2026 12:00:00 GMT</pubDate><description>How fill-in-the-middle (FIM) training lets code models use the code above and below your cursor, from the 2022 paper to Code Llama, StarCoder2 and Ollama's suffix.</description></item>
<item><title>MCP's old initialize handshake, and why it was retired</title><link>https://ai.miladezzat.com/mcp-11-initialize-handshake/</link><guid>https://ai.miladezzat.com/mcp-11-initialize-handshake/</guid><pubDate>Sat, 14 Nov 2026 12:00:00 GMT</pubDate><description>How legacy MCP connections opened with initialize, negotiated a version and capabilities, and what the stateless 2026-07-28 revision does instead.</description></item>
<item><title>Prompt versioning: ship prompts like code</title><link>https://ai.miladezzat.com/prompt-versioning/</link><guid>https://ai.miladezzat.com/prompt-versioning/</guid><pubDate>Sat, 14 Nov 2026 12:00:00 GMT</pubDate><description>Track every prompt change as a version, ship it by moving a production label, roll back in seconds, and compare versions in traces and evals.</description></item>
<item><title>LLM API gateways: one door to every model</title><link>https://ai.miladezzat.com/llm-api-gateways/</link><guid>https://ai.miladezzat.com/llm-api-gateways/</guid><pubDate>Fri, 13 Nov 2026 12:00:00 GMT</pubDate><description>What an LLM gateway between your apps and model providers gives you: one endpoint, virtual keys and budgets, cost tracking, retries, fallbacks and caching.</description></item>
<item><title>MCP Inspector: watch your MCP server's protocol live</title><link>https://ai.miladezzat.com/mcp-10-inspector/</link><guid>https://ai.miladezzat.com/mcp-10-inspector/</guid><pubDate>Fri, 13 Nov 2026 12:00:00 GMT</pubDate><description>How to run the MCP Inspector against a local or remote server, call tools from a form, read the JSON-RPC transcript, and switch between legacy and modern protocol eras.</description></item>
<item><title>PII redaction for LLM apps: mask it in, restore it out</title><link>https://ai.miladezzat.com/pii-redaction/</link><guid>https://ai.miladezzat.com/pii-redaction/</guid><pubDate>Fri, 13 Nov 2026 12:00:00 GMT</pubDate><description>Keep names, emails and card numbers out of your prompts. Detect personal data, swap it for placeholders, keep the mapping, and restore the real values in the answer.</description></item>
<item><title>SWE-bench explained: how coding agents get graded</title><link>https://ai.miladezzat.com/swe-bench/</link><guid>https://ai.miladezzat.com/swe-bench/</guid><pubDate>Fri, 13 Nov 2026 12:00:00 GMT</pubDate><description>What a SWE-bench score really measures, why SWE-bench Verified exists, and why contamination led to SWE-Bench Pro. How to read a coding benchmark score.</description></item>
<item><title>Add your first MCP server to Claude, Claude Code, Cursor or VS Code</title><link>https://ai.miladezzat.com/mcp-09-first-server/</link><guid>https://ai.miladezzat.com/mcp-09-first-server/</guid><pubDate>Thu, 12 Nov 2026 12:00:00 GMT</pubDate><description>The one config shape every MCP client shares, and exactly where it goes in Claude Desktop, Claude Code, Cursor and VS Code, with the one key name that differs.</description></item>
<item><title>Long context vs RAG: paste it all, or retrieve it?</title><link>https://ai.miladezzat.com/long-context-vs-rag/</link><guid>https://ai.miladezzat.com/long-context-vs-rag/</guid><pubDate>Thu, 12 Nov 2026 12:00:00 GMT</pubDate><description>Million-token context windows can hold thousands of pages. When to paste everything into the prompt, when to retrieve only what matches, and how to combine both.</description></item>
<item><title>Text-to-SQL: ask your database in plain English</title><link>https://ai.miladezzat.com/text-to-sql/</link><guid>https://ai.miladezzat.com/text-to-sql/</guid><pubDate>Thu, 12 Nov 2026 12:00:00 GMT</pubDate><description>Text-to-SQL lets an LLM turn a plain-English question into a SQL query. How it works, what the BIRD and Spider benchmarks show, and how to run it safely.</description></item>
<item><title>Top-p vs top-k sampling: what the knobs actually cut</title><link>https://ai.miladezzat.com/top-p-top-k-sampling/</link><guid>https://ai.miladezzat.com/top-p-top-k-sampling/</guid><pubDate>Thu, 12 Nov 2026 12:00:00 GMT</pubDate><description>Top-k, top-p (nucleus) and min-p sampling all cut the long tail of unlikely tokens. How each one decides where to cut, and why a fixed k fits badly.</description></item>
<item><title>AI code review: a model on your pull requests</title><link>https://ai.miladezzat.com/ai-code-review/</link><guid>https://ai.miladezzat.com/ai-code-review/</guid><pubDate>Wed, 11 Nov 2026 12:00:00 GMT</pubDate><description>AI code review tools read your diff and comment on the changed lines. What they catch, how Copilot and Claude fit in, what Google measured, and why a person still decides.</description></item>
<item><title>Hybrid search: BM25 plus vectors, merged with RRF</title><link>https://ai.miladezzat.com/hybrid-search/</link><guid>https://ai.miladezzat.com/hybrid-search/</guid><pubDate>Wed, 11 Nov 2026 12:00:00 GMT</pubDate><description>Vector search understands meaning but misses exact strings; BM25 does the opposite. Hybrid search runs both and merges them with Reciprocal Rank Fusion.</description></item>
<item><title>Synthetic data: training AI on data a model wrote</title><link>https://ai.miladezzat.com/synthetic-data/</link><guid>https://ai.miladezzat.com/synthetic-data/</guid><pubDate>Wed, 11 Nov 2026 12:00:00 GMT</pubDate><description>Synthetic data is training and test data generated by a model. How Self-Instruct, Alpaca and phi-1 used it, and why model collapse means you still need real data.</description></item>
<item><title>Who runs MCP? Governance, SEPs and the deprecation policy</title><link>https://ai.miladezzat.com/mcp-08-governance/</link><guid>https://ai.miladezzat.com/mcp-08-governance/</guid><pubDate>Wed, 11 Nov 2026 12:00:00 GMT</pubDate><description>MCP lives at the Agentic AI Foundation. How its maintainers decide, how a SEP moves from draft to final, and how long deprecated features stay in the spec.</description></item>
<item><title>Batch APIs: cut your LLM bill in half by waiting</title><link>https://ai.miladezzat.com/batch-apis/</link><guid>https://ai.miladezzat.com/batch-apis/</guid><pubDate>Tue, 10 Nov 2026 12:00:00 GMT</pubDate><description>Batch APIs run LLM requests in the background at 50% of the normal price. How they work, their limits, and which jobs to move to a batch.</description></item>
<item><title>Diffusion models: how AI turns static into a picture</title><link>https://ai.miladezzat.com/diffusion-models/</link><guid>https://ai.miladezzat.com/diffusion-models/</guid><pubDate>Tue, 10 Nov 2026 12:00:00 GMT</pubDate><description>Diffusion models learn to remove noise, then generate images by starting from pure static and denoising step by step. How training, sampling and guidance work.</description></item>
<item><title>Semantic caching: same meaning, same answer</title><link>https://ai.miladezzat.com/semantic-caching/</link><guid>https://ai.miladezzat.com/semantic-caching/</guid><pubDate>Tue, 10 Nov 2026 12:00:00 GMT</pubDate><description>Users ask the same question a hundred ways. Semantic caching matches questions by meaning, returns the saved answer, and skips the LLM call. How to tune it safely.</description></item>
<item><title>stdio vs Streamable HTTP: the two MCP transports</title><link>https://ai.miladezzat.com/mcp-07-transports/</link><guid>https://ai.miladezzat.com/mcp-07-transports/</guid><pubDate>Tue, 10 Nov 2026 12:00:00 GMT</pubDate><description>MCP messages travel over stdio for local servers or Streamable HTTP for remote ones. How each works in the 2026-07-28 spec, and the rules to get right.</description></item>
<item><title>How MCP is versioned</title><link>https://ai.miladezzat.com/mcp-06-versioning/</link><guid>https://ai.miladezzat.com/mcp-06-versioning/</guid><pubDate>Mon, 09 Nov 2026 12:00:00 GMT</pubDate><description>MCP versions aren't 1.0 or 2.0, they're dates. What each version from 2024-11-05 to 2026-07-28 changed, and how clients and servers agree on one. Episode 6.</description></item>
<item><title>RLHF: reinforcement learning from human feedback</title><link>https://ai.miladezzat.com/rlhf/</link><guid>https://ai.miladezzat.com/rlhf/</guid><pubDate>Mon, 09 Nov 2026 12:00:00 GMT</pubDate><description>A small model beat one a hundred times bigger, thanks to human feedback. How RLHF works step by step, why it can cause sycophancy, and newer methods like DPO.</description></item>
<item><title>Speculative decoding: faster LLMs, same output</title><link>https://ai.miladezzat.com/speculative-decoding/</link><guid>https://ai.miladezzat.com/speculative-decoding/</guid><pubDate>Mon, 09 Nov 2026 12:00:00 GMT</pubDate><description>A small model can make a big one faster with exactly the same output. How speculative decoding drafts tokens and verifies them in one pass, and when it helps.</description></item>
<item><title>The KV cache: why the first token is slow</title><link>https://ai.miladezzat.com/kv-cache/</link><guid>https://ai.miladezzat.com/kv-cache/</guid><pubDate>Mon, 09 Nov 2026 12:00:00 GMT</pubDate><description>Why is the first token slow but the rest stream out fast? The KV cache. How prefill and decode work, how much memory the cache uses, and how PagedAttention helps.</description></item>
<item><title>JSON-RPC 2.0 in MCP</title><link>https://ai.miladezzat.com/mcp-05-json-rpc/</link><guid>https://ai.miladezzat.com/mcp-05-json-rpc/</guid><pubDate>Sun, 08 Nov 2026 12:00:00 GMT</pubDate><description>Every MCP message is plain JSON-RPC 2.0. Requests, responses and notifications, ids, error codes and resultType, explained. Episode 5 of MCP from A to Z.</description></item>
<item><title>RAG chunking strategies</title><link>https://ai.miladezzat.com/rag-chunking/</link><guid>https://ai.miladezzat.com/rag-chunking/</guid><pubDate>Sun, 08 Nov 2026 12:00:00 GMT</pubDate><description>Your RAG app has the answer but keeps retrieving the wrong paragraph. Often the problem is chunking. Fixed-size, recursive, semantic and contextual chunking compared.</description></item>
<item><title>Rerankers: put the best result first</title><link>https://ai.miladezzat.com/rerankers/</link><guid>https://ai.miladezzat.com/rerankers/</guid><pubDate>Sun, 08 Nov 2026 12:00:00 GMT</pubDate><description>Your search found the right document but ranked it ninth. Rerankers read the query and each document together to reorder results. Bi-encoders vs cross-encoders explained.</description></item>
<item><title>Voice agents: AI you talk to</title><link>https://ai.miladezzat.com/voice-agents/</link><guid>https://ai.miladezzat.com/voice-agents/</guid><pubDate>Sun, 08 Nov 2026 12:00:00 GMT</pubDate><description>Two seconds of silence and a voice bot already feels broken. How voice agents work: chained vs speech-to-speech, turn detection, barge-in and latency.</description></item>
<item><title>Agent sandboxing and permissions</title><link>https://ai.miladezzat.com/agent-sandboxing/</link><guid>https://ai.miladezzat.com/agent-sandboxing/</guid><pubDate>Sat, 07 Nov 2026 12:00:00 GMT</pubDate><description>Your coding agent ran a command you never read. What stopped it from wiping your home folder? Permissions decide what it may do; a sandbox limits what it can do.</description></item>
<item><title>Background coding agents</title><link>https://ai.miladezzat.com/background-coding-agents/</link><guid>https://ai.miladezzat.com/background-coding-agents/</guid><pubDate>Sat, 07 Nov 2026 12:00:00 GMT</pubDate><description>Assign a ticket before lunch, and a pull request is waiting when you're back. How background coding agents work in cloud sandboxes, and how to use them well.</description></item>
<item><title>C2PA and AI watermarks</title><link>https://ai.miladezzat.com/c2pa-ai-watermarks/</link><guid>https://ai.miladezzat.com/c2pa-ai-watermarks/</guid><pubDate>Sat, 07 Nov 2026 12:00:00 GMT</pubDate><description>Is this photo real, or did AI make it? Content credentials record where a file came from; watermarks hide a signal inside it. How C2PA and SynthID work, and their limits.</description></item>
</channel></rss>
