AI Ustoz UZ: learn artificial intelligence in Uzbek

Formats atlas · cheat sheet

File formats and digital hygiene: a cheat sheet

The main takeaways of the Formats atlas video series on one page: 10 protection rules, how risky each format is and the key points of every part.

Download PDF (in Uzbek)2 pages, A4, in Uzbek

10 digital hygiene rules

  1. Make file extensions visible

    In Windows: File Explorer → View → File name extensions. Otherwise the program “report.pdf.exe” shows up as “report.pdf”.

  2. Keep Protected View on

    Do not turn Protected View off; click “Enable Editing” only for files from a source you trust.

  3. Do not enable macros

    Do not “Unblock” a file that came from the internet: Office on Windows blocks macros in such files by default.

  4. Confirm an unexpected file through another channel

    If an unexpected file claims to be from an organisation, phone them or go to their official website yourself.

  5. Never upload confidential files to online scanners

    Scan a suspicious file for viruses, but do not upload confidential documents or patient data to an online service such as VirusTotal.

  6. Open unknown files in an isolated environment

    Windows Sandbox or a virtual machine leaves your main system untouched.

  7. Keep your system and apps updated

    System, browser, PDF reader and Office: even when a patch exists, a system that is not updated stays vulnerable.

  8. Be careful with password-protected archives

    A password sent in the same message is a warning sign.

  9. Download software only from official sources

    Use the official store or the official website, and compare the hash or signature of the downloaded file.

  10. Turn on two-step protection

    Choose a phishing-resistant method where possible. According to a Microsoft study, multi-factor authentication lowers the risk of account compromise by about 99%.

Format → risk level

The author's personal assessment, not an official rating; based on vendor documentation.

FormatRisk levelWhat to do
.txt .mdlowPlain text; check that the name is not “file.txt.exe”.
.jpg .png .webp .mp4 .mp3lowImage and media libraries have had vulnerabilities (WebP, CVE-2023-4863): keep updated.
.csvlow to mediumCan run as a formula in Excel (OWASP); open files from outside in a text editor.
.doc .docx .xls .xlsx .ppt .pptxlow to mediumOpen in Protected View; do not trust a request to “Enable Editing”.
.pdfmediumOpen in Protected View / Enhanced Security; be wary of links and “sign in” prompts.
.ipa · TestFlight · EU marketplacesmediumApple review exists, but it is not safer than the official App Store.
.html .htm .svg (attachment)medium to highMay be a fake sign-in page; check the address.
.zip .rar .iso .img (especially password-protected)high / mediumAn executable can hide inside; look at the contents first.
.docm .xlsm .pptmhighMacros; Office on Windows blocks them by default for internet files; do not “Unblock”.
.exe .msi .bat .ps1 .vbs .js .scr .lnkhighRuns with full rights when launched.
.apk (outside Google Play)highInstalling outside Google Play (sideloading): check the source, signature and permissions.

The four parts of the series: key takeaways

Part 1

Text and data

  • Text and data formats (Markdown, CSV, JSON, Python) are the easiest for artificial intelligence: simple, short, precise.
  • The same 30-row table: 525 tokens as clean text versus 9,493 tokens in the internal XML of a Word (docx) file — about 18 times more.
  • Rule: keep working files in plain text, make the final deliverable pretty.
  • If a CSV cell starts with =, +, - or @, Excel reads it as a formula (OWASP: CSV injection).
  • Reading a script (.bat, .ps1, .js, .vbs, .py) is safe; running one from a stranger is not.
Watch the video

Part 2

Documents and office

  • Word, Excel and PDF look good to people but are complex for machines: a .docx file is really XML inside a ZIP archive.
  • Convert to text or Markdown before giving a file to artificial intelligence: free local tools such as MarkItDown (Microsoft), Docling (IBM) and Pandoc.
  • Ready-made tools make mistakes too (a two-column PDF came out jumbled): always review the result.
  • Never upload a document with a patient’s name, ID or diagnosis to an online converter or a general chat app: use a local tool or fully anonymise it first.
  • A PDF “from an organisation” with an unexpected sender, urgency, a “sign in” link or a QR code: confirm it through another channel.
Watch the video

Part 3

Images, audio and video

  • Claude counts an image in 28×28-pixel blocks: a 1000×1000 image is 1,296 tokens; a 4K image is 4,784 tokens in high-resolution mode and 1,560 in standard mode.
  • If there is text, give text — do not screenshot a table; shrink images you do not need at full size.
  • Convert iPhone photos (HEIC) to JPEG or PNG; from an animated GIF Claude takes only the first frame.
  • MP4 is a container: the codec inside varies, so the same file does not open everywhere.
  • For audio and video, transcribe or extract frames first; limits change quickly, so check your model’s documentation.
Watch the video

Part 4

Cybersecurity

  • The danger is not in the format’s name but in what is inside and how you open it: even an image can be risky.
  • A common trick: “report.pdf.exe”. If Windows hides extensions, you only see “report.pdf”.
  • On phones the biggest risk is installing apps outside Google Play. Check the source, the developer and the permissions.
  • Packages and VS Code extensions: check the name letter by letter, look at the publisher and install count; open unknown folders in Restricted Mode.
  • Check suspicious files locally (hash, Defender, Sandbox); never upload confidential files to online services.
Watch the video

Bottom line: it is habit and a checking routine that protect you, not the format. Pause before opening, check the source, keep things updated, never upload confidential data.

Information as of October 2026. Token counts use the OpenAI tokenizer (o200k) and Claude documentation; other models give different numbers. Unverified information is not included. The series author is a physician. The videos are in Uzbek.