How to Resize Images for Social Media: Complete Guide 2026
Master the exact image dimensions for Instagram, Facebook, Twitter, LinkedIn, and YouTube. Learn optimization techniques, aspect ratios, and best practices for maximum engagement on every platform.
Tokens vs Words vs Characters: The Conversion Table
Working conversion ratios between words, characters and tokens — where the familiar rule of thumb holds, where it fails by a factor of three, and how to estimate without underrunning your context window.
Why LLMs Can't Count the R's in "Strawberry" — Tokenisation Explained
A model that writes working code fails at counting letters in a word. It is not a reasoning failure — the model never receives letters in the first place. What tokenisation hides, and which tasks it quietly breaks.
How to Estimate LLM API Costs Before You Run Anything
The formula for costing an LLM feature before you build it — plus the four multipliers that turn a careful estimate into a bill four times larger, and why chat costs grow quadratically.
Where Vectors Live: Parquet, NPY and Vector Database Storage
A million embeddings stored as JSON is roughly 20GB. The same vectors as float16 in Parquet fit in under a gigabyte. How embeddings are stored, and when you actually need a vector database.
YAML vs JSON for Prompt and Agent Configuration
Prompts are multiline text, which JSON handles badly and YAML handles well. But YAML will also decide your country code is a boolean. Which format belongs where, and how to avoid the traps in both.
Token-Efficient Data Formats: JSON vs YAML vs CSV vs Markdown Tables
The same data can cost three times as many tokens depending on how you format it. Where the overhead actually hides, what the savings look like at real sizes, and the cases where the cheapest format is the wrong choice.
ONNX Explained: One Model File, Every Runtime
Train in PyTorch, deploy to a phone, a browser or a C# service without shipping the framework. What an .onnx file contains, and why GGUF took over local LLMs regardless.
How Big Is a Model File? Parameters → Gigabytes → VRAM
Parameters times bytes per parameter gives the file size. Running it needs more — and at long context the KV cache, not the weights, is what exhausts your memory.
What's Inside a .safetensors File, Byte by Byte
Eight bytes of length, a JSON header, then raw tensor bytes. A deliberately boring format — and the boringness is what makes it memory-mappable, inspectable and impossible to execute.
LoRA Files: Why Adapters Are 100× Smaller Than Models
A full fine-tune of a 7B model is 14GB. The same adaptation as a LoRA adapter is often under 20MB. The matrix decomposition that makes that work, and what it cannot do.
Running an LLM Locally: Real File and Hardware Requirements
Memory bandwidth divided by model size predicts your token rate better than any spec sheet. What local inference really needs, which quantisation to pick, and when it beats an API.
How to Tell If an Image Was AI-Generated
Metadata first, visual tells second, detectors a distant last. An honest account of what works, what does not, and why a confident detection score is the weakest evidence you can have.
Why AI Tools Output PNG — And Why You Should Ship WebP
Generators output PNG for good reasons that stop applying the moment you publish. Why a 1.8MB generated image should reach your users as 180KB, and what to keep in the archive.
What Is a Seed in AI Image Generation?
A seed does not store a picture. It sets the initial noise, and the image emerges from that noise plus every other setting — which is why the same seed on a different model means nothing at all.
Upscaling AI Images: What Works and What Just Adds Pixels
No upscaler recovers lost detail. One interpolates honestly, the other invents convincingly — and the difference decides whether the output is usable as art, as evidence, or at all.
Stripping Metadata From AI-Generated Images
A generated PNG can carry your full prompt, negative prompt, seed, model hash and an entire node workflow. How to remove it, when you should not, and how to verify it actually worked.
Why AI Images Have Artefacts: Latent Space and VAE Explained
Diffusion does not happen at the resolution you see. It happens in a compressed latent space roughly 8x smaller per side, and a decoder expands the result. Most artefacts follow directly from that.
What Actually Happens to Files You Upload to AI Tools
"We don't train on your data" answers one question out of five. What actually happens to an uploaded file — the tier you're on, the retention you didn't agree to, and what deletion cannot undo.
Indirect Prompt Injection: When Your Own Documents Attack You
The attacker never touches your system. They plant instructions in a document, a page or an email your assistant will later read, and your own user triggers it with their permissions.
How to Redact a Document Before Sending It to an AI
Drawing a black box over text does not remove it — and an AI reads the text layer, not the picture. Where content actually hides, and why extracting what you need beats redacting what you don't.
Browser-Based AI vs Cloud AI: The Real Privacy Trade-off
One is a promise you have to trust; the other is a property you can verify in the network tab. What browser-based AI genuinely provides, what it costs, and why the hybrid split usually wins.
Uploading Data to AI Tools: The GDPR Position
Controller or processor, lawful basis, transfers and minimisation — and the genuinely unresolved problem of a right to erasure that meets model weights nobody can un-train.
Metadata Is a Bigger AI Privacy Risk Than Your Content
You redacted the names in the text. The author field, the file path, the GPS coordinates and the revision history went anyway — and unlike prose, metadata is structured, precise and trivially aggregated.
What Is Agentic AI? The Definitions Nobody Agrees On
Four incompatible definitions of agent are in circulation, and vendors move between them freely. The properties that actually distinguish agentic systems, and the questions that cut through the marketing.
MCP Explained: The Protocol That Standardised AI Tool Use
Every AI application used to write its own integration for every tool — N applications times M tools, all bespoke. MCP turns that into N plus M. How the protocol works, and what you accept when you install a server.
Tool Calling: How an LLM Actually Uses a Tool
The model never runs your code. It emits a structured request, your code decides whether to honour it, and the result goes back as text. Every security property of tool use follows from that gap.
Agents vs Workflows: When You Don't Need an Agent
A workflow follows a path you wrote. An agent decides its own. Most problems described as needing an agent are workflows in disguise, and building them as workflows makes them cheaper, faster and debuggable.
How AI Agents Read and Write Files
Files are the most common thing agents touch and the easiest to damage. Why reads are chunked, why whole-file rewrites lose content, and the containment that actually works.
Building a Document-Processing Agent That Actually Works
A worked design for pulling structured data out of mixed documents: where the model belongs, why most of the pipeline should not be agentic, and how to handle the exceptions that genuinely are.
Why AI Agents Fail: Context Rot, Loops and Error Handling
Agents rarely fail loudly. They drift off task, loop on a broken step, or announce success without doing the work. The six real failure modes, why long runs degrade, and what to instrument.
llms.txt and robots.txt: Controlling AI Crawlers on Your Site
One of these files is a thirty-year-old standard that AI companies mostly honour. The other is a 2024 proposal that almost nothing reads. Which crawlers to allow, which to block, and why the most common blocking advice backfires.
Social Media Image Size Cheat Sheet: Every Platform in 2026
Exact pixel dimensions for every major platform in one table — feed posts, stories, covers, thumbnails and profile pictures — plus the safe zones and compression rules nobody documents.
HTTP Status Codes: The Complete Reference
Every status code in one table, plus the distinctions that trip up experienced developers: 301 vs 308, 302 vs 307, 401 vs 403, and why 422 exists at all.
MIME Types Explained (And Why Your Upload Keeps Getting Rejected)
The full MIME type list for every common format — plus how browsers really decide a file's type, and why renaming a .webp to .jpg never fixes anything.
File Extension Reference: What Opens Every Common File Type
Look up any extension: what the format actually is, what opens it, and which ones can run code the moment you double-click them.
Paper Sizes in Pixels: A4, Letter and Every ISO Size at Any DPI
A4, A3, Letter and the rest in pixels at 72, 96, 150, 300 and 600 DPI — plus the elegant maths that makes the A series work and why 300 DPI became the print standard.
Photo Print Sizes: The Complete Chart (And Why Your Photo Gets Cropped)
Every standard print size with the pixels and megapixels it needs — plus why an 8×10 always crops your photo and a 4×6 never does.
All 148 CSS Color Names With Hex and RGB Values
Every CSS colour keyword with exact hex and RGB values and live swatches — plus the strange history behind names like papayawhip, and why nine of them are duplicates.
Screen Resolution Chart: 720p to 8K, and What CSS Pixels Really Are
Every standard resolution from 720p to 8K, plus the thing that confuses everyone: why a phone with 1179 physical pixels tells your CSS it is 393 pixels wide.
Aspect Ratio Chart: Every Common Ratio, With the Maths Behind It
Every ratio from 1:1 to 32:9 with matching resolutions, the arithmetic to work out any dimension yourself, and why black bars appear where they do.
Font Size Conversion: px, pt, em, rem, % and When to Use Each
The full conversion table between px, pt, em, rem and percent — plus the compounding trap in em and the accessibility reason your unit choice matters more than you think.
Date Format Reference: ISO 8601, RFC 3339 and Every strftime Code
Every strftime code in one table, the real difference between ISO 8601 and RFC 3339, and why 03/04/2026 means two different days depending on who reads it.
Cron Syntax Reference: Every Field, Operator and 40 Working Examples
Every field and operator, 40 copy-ready schedules, and the day-of-month/day-of-week rule that makes some cron jobs run every single day instead of once a month.
Regex Syntax Reference: Where JavaScript, Python, PCRE and Go Disagree
Every regex token in one table, plus the specific places JavaScript, Python, PCRE and Go disagree — and the pattern shape that can hang a server.
Unicode Character Reference: The Symbols You Actually Need
Dashes, quotes, arrows, maths and currency symbols with code points and entities — plus the invisible characters that silently break comparisons and the reason an emoji has a length of 2.
Email Attachment Size Limits — And Why 25MB Really Means 18MB
Provider-by-provider limits, plus the encoding overhead nobody mentions: a 25MB cap only fits an 18MB file, and the recipient's server has a veto.
Why Windows Says a File Is 'In Use' — And How to Actually Fix It
Find the exact process holding your file, understand why Windows locks files and Linux never does, and delete it safely without the reboot.
Why Your Photo Rotates Itself: EXIF Orientation Explained
The same photo, upright in one app and sideways in another. Here is the metadata flag behind it, why rotating does not stick, and how to fix it without re-compressing.
Why Your PDF Prints Differently Than It Looks on Screen
Fonts substituted, colours dulled, edges cropped, transparency turned into grey boxes — the real causes of PDF print differences and how to identify which one is yours.
Why Your Fonts Change on Someone Else's Computer
Why a document looks different on every machine, how substitution silently changes your layout rather than just your styling, and the reliable ways to make fonts travel.
Why Excel Keeps Changing Your Data — And How to Stop It
Leading zeros vanish, long IDs become 1.23457E+15, and product codes turn into dates. The real mechanism behind each one, and how to import without damage.
Why Your ZIP File Won't Open — Every Cause and Fix
Incomplete downloads, AES encryption Windows cannot read, mangled filenames, path limits and split archives — how to tell which one broke your ZIP, and how to recover it.
Why Emojis Show as Empty Boxes — And How to Fix It
The empty box has a name — tofu — and it always means one thing. Here is what causes it, why Windows refuses to show flag emoji, and how compound emoji fall apart.
Why Your Browser Says a Download Is Unsafe — And When to Believe It
Four separate systems produce these warnings and they mean different things. Here is what each one is actually claiming, and how to verify a download rather than guessing.
Why Your 1TB Drive Shows Only 931GB
The exact arithmetic behind the missing 69GB, why the manufacturer is technically correct, and where the remaining space genuinely goes.
Why the Same File Shows Different Sizes in Different Places
Size versus size on disk, cluster slack, compression, sparse files and hard links — the six reasons two tools report different numbers for one file.
Why Your Scanned Documents Are Enormous — And How to Fix It
The same page can scan to 50KB or 25MB. Three settings decide which — here is the arithmetic, and how to shrink scans you already have.
Why Your Screenshots Look Fuzzy — And How to Get Sharp Ones
Sharp on your screen, soft everywhere else. The four causes — 2x capture, fractional scaling, JPEG on text, and non-integer resizing — and how to avoid each.
Why Your Filenames Turn Into Question Marks and Strange Symbols
Question marks, mojibake and boxes each mean something different — one is recoverable and one is permanent. Here is how to tell which you have, and repair a whole folder.
Why Your Video Won't Play — Containers vs Codecs Explained
Two files both called .mp4, one plays and one does not. The container is the box and the codec is the contents — here is how to find out what is inside and fix it in seconds.
Why Copy-Paste Brings Weird Formatting — And How to Stop It
Your clipboard holds the same content in five formats at once, and the receiving app silently picks the richest one. Here is how to see what is really there and control it.
How Image Compression Actually Works
What actually happens inside JPEG, PNG and AVIF — why JPEG destroys text, why PNG is enormous for photographs, and what the quality slider really adjusts.
How a QR Code Actually Encodes Data (Decode One by Hand)
Every region of a QR code has a job. Here is what each one does, why the pattern looks random, and how to read the version and error correction level by eye.
What Actually Happens When You Delete a File
Deleting removes a pointer, not the data. Here is what actually survives, why SSDs changed the rules completely, and why encryption is the only erasure that works.
Why Every System Picked a Different Zero Date for Time
Windows counts from 1601, Excel from 1900, Macs from 1904, GPS from 1980. Each choice had a reason — and each left behind a bug that is still with us.
Why Is US Letter 8.5 × 11 Inches?
Nobody wrote down why. The competing origin stories, the 1980 decision that made it official, and the engineering cost of a size series that cannot scale cleanly.
How Unicode Saved the Internet From Itself
Before Unicode, the same bytes meant different letters in every country and no document could say which. How that got fixed — including the design sketched on a diner placemat.
How a Computer Stores a Colour — And Why 128 Is Not Half Brightness
RGB 128 emits about 21% of the light of 255, not 50%. That one fact explains why resized images come out too dark and why blending colours the obvious way is wrong.
The History of the ZIP File — A Lawsuit, a Format, and a Tragedy
Every ZIP file starts with the letters PK. They are a man's initials, and the format exists because he was sued — a decision that made ZIP the container inside DOCX, APK, JAR and EPUB.
Why PNG Exists — The GIF Patent War That Created It
In 1994 a patent holder decided to start charging for GIF. The web designed a replacement in weeks — and GIF survived anyway. Here is why both things happened.
Anatomy of a PNG File, Byte by Byte
Read a PNG by hand — the signature, the chunk layout, and the rule where the capitalisation of each letter in a chunk name is itself machine-readable information.
Anatomy of a JPEG File: Markers, Segments and Hidden Data
The marker system, every segment type, where EXIF really lives — and the embedded thumbnail that can still show what you cropped out.
What's Actually Inside a .docx File
A Word document is a ZIP of XML. What every file inside does, why searching the text so often fails, and what your document is still carrying from earlier drafts.
How PDF Files Are Structured Internally
The object model, the cross-reference table, and the incremental update mechanism that quietly keeps every earlier version of the document inside the file.
Magic Numbers: How Software Identifies Files Without Extensions
The byte signatures every format begins with, why they beat extensions and MIME types, and how to validate an upload that is lying about what it is.
What Is a Token in an LLM? (And Why Your Bill Doesn't Match Your Word Count)
Tokens are not words. Here is how tokenisation really works, why the same sentence costs triple in some languages, and how to estimate your bill before you spend anything.
What a Context Window Really Is — And Why Bigger Isn't Better
Advertised context and usable context are different numbers. Why attention costs scale quadratically, what lost-in-the-middle means for your retrieval, and how to budget a window.
Chunking Text for RAG: Sizes, Overlap and Where to Split
Chunk size is a precision-versus-context trade-off with no universal answer. Which strategy suits which document, how to handle tables, and the technique that fixes most retrieval failures.
Why PDFs Break RAG — And How to Prepare Them Properly
A PDF has no paragraphs, no reading order and no real tables — only positioned marks. Here is exactly how that wrecks retrieval, and the extraction pipeline that fixes it.
Why Markdown Is the Best Format to Feed an LLM
The same content as HTML costs 4× the tokens of Markdown and conveys less. The measured comparison, why models handle Markdown so well, and the conversion pipeline.
Structured Outputs: Getting Valid JSON From an LLM Every Time
Constrained decoding makes invalid JSON structurally impossible — here is the mechanism, what a schema still cannot guarantee, and how to design one the model actually gets right.
Tool Schemas: Designing Functions an LLM Can Actually Use
Tool definitions are resent on every single request and are usually the largest hidden cost in an agent. How to design ones a model picks correctly — and how to keep them cheap.
Preparing a CSV for LLM Analysis Without Sending 100,000 Rows
You cannot paste 100,000 rows into a prompt. The three strategies that work, plus the formatting details — rounding, column pruning, date normalisation — that halve your token bill.
What Is an Embedding? Vectors Explained Without the Maths
What a vector actually encodes, why 'similar' is not the same as 'relevant', and the storage maths — plus the quantisation tricks that cut a 6GB index to 200MB.
JSONL and Fine-Tuning Datasets: Format, Size and Quality
Why every fine-tuning API uses JSONL, how many examples you actually need, and the quiet mistakes — duplicates, leakage, inconsistent formatting — that waste a training run.
Prompt Caching: How It Works and What It Actually Saves
Caching can cut input cost by up to 90% and latency along with it — but only for a byte-identical prefix. Here is how to restructure a prompt so it actually applies.
GGUF Explained: The File Format Behind Local LLMs
What is actually inside a GGUF file, why memory mapping lets you load a 40GB model instantly, and how to decode quantisation names like Q4_K_M.
Safetensors vs Pickle: Why Model File Format Is a Security Problem
Loading a .pt model can execute arbitrary code on your machine before a single weight is read. The mechanism, why scanners cannot fully fix it, and what safetensors changed.
What Model Quantisation Actually Does
How precision reduction shrinks a model, why '4-bit' is really 4.8 bits, what perplexity hides, and why a bigger model at Q4 beats a smaller one at Q8.
AI Image Generators Hide Your Prompt Inside the File
Your full prompt, negative prompt, seed, model hash and settings are written into the PNG. ComfyUI embeds the entire node graph. Here is where it lives and how to remove it.
C2PA and Content Credentials: How Image Provenance Actually Works
Signed manifests can prove an image's origin and edit history — until someone screenshots it. What C2PA actually guarantees, and what it cannot.
Is It Safe to Upload Documents to AI Tools?
What actually happens to an uploaded file — training use, retention windows, the metadata you send without realising, and a practical rule for deciding per document.
Prompt Injection Explained — And Why It Isn't Solved
An LLM cannot tell instructions from data — that is the whole vulnerability, and it is not a bug to be patched. Why filtering fails and what architectural controls actually help.
JPG vs PNG: Which Image Format Should You Use?
Understand the key differences between JPG and PNG formats. Learn when to use each, how compression works, and make the right choice for your projects.
How to Merge PDFs Without Adobe Acrobat
Discover free methods to combine PDF files without expensive software. Step-by-step guides for Windows, Mac, and online tools.
QR Codes Explained: Everything You Need to Know
From history to modern applications, learn how QR codes work, different types, best practices for creation, and creative ways businesses use them.
JSON for Beginners: What It Is and How to Use It
A comprehensive introduction to JSON (JavaScript Object Notation). Learn syntax, data types, common use cases, and how to work with JSON in your projects.
How to Compress Images Without Losing Quality
Learn the science behind image compression. Discover techniques to reduce file sizes by 50-80% while maintaining visual quality for web and print.
Best Image Formats for Web: WebP, AVIF, JPG & PNG
Compare modern image formats for web use. Learn when to use WebP, AVIF, JPG, and PNG for optimal performance and quality.
How to Create a PDF from Multiple Images
Step-by-step guide to combining photos into a single PDF. Free methods for Windows, Mac, iPhone, Android, and online tools.
QR Code Types Explained: Static vs Dynamic & More
Complete guide to QR code types: URL, vCard, WiFi, Email, SMS, Location. Learn static vs dynamic codes and when to use each.
File Security Best Practices: Protect Your Documents
Essential file security guide: passwords, encryption, secure sharing, backups, and protection against threats. Complete 2026 guide.
Word Counter Tools: The Complete Guide to Counting Words & Characters
How word and character counts really work, why different tools give different numbers, and how to hit exact limits for essays, Twitter/X, meta descriptions, and more.
Lorem Ipsum: History, Purpose & How to Use Placeholder Text
Where lorem ipsum actually comes from (Cicero, 45 BC), why designers still use scrambled Latin, and when real placeholder text is the better choice.
Regular Expressions for Beginners: A Complete Guide
Learn regex from zero: character classes, quantifiers, anchors, groups, and real patterns for validating emails, phone numbers, and dates — with worked examples.
Unix Timestamps Explained: What They Are & How to Convert Them
Why computers count seconds since January 1, 1970, how time zones and leap seconds fit in, and how to convert timestamps in JavaScript, Python, SQL, and Excel.
JSON to CSV Conversion: Complete Guide for Data Processing
How to flatten nested JSON into spreadsheet-ready CSV, handle arrays and missing fields, avoid Excel encoding pitfalls, and automate conversions.
SVG vs PNG: When and How to Convert Between Vector and Raster
Understand the real difference between vector and raster graphics, when each format wins, and how to convert SVG to crisp PNG at any resolution.
Color Codes Explained: HEX, RGB, HSL Complete Guide
What #FF6B35 actually means, how RGB and HSL describe the same color differently, and which notation to use in CSS, design tools, and print work.
Color Theory for Web Design: A Complete Beginner's Guide
Build color palettes that actually work: the color wheel, harmony rules, the 60-30-10 rule, accessibility contrast requirements, and common beginner mistakes.
How to Extract Colors from Images: Guide to Color Palettes
Turn any photo into a usable color palette: how dominant-color algorithms work, building brand palettes from product shots, and applying extracted colors in real designs.