About Packet To Sniff
Packet To Sniff teaches how AI systems and networks really work: tokens, context windows, retrieval, machine learning methods, and the security problems they create. Plain-language articles, browser tools, hands-on labs and a tutor that shows its sources.
Why this site exists
Most material about AI either stays at the level of marketing or jumps straight to research papers. Packet To Sniff sits in between: it explains what actually happens to your text inside a language model, how retrieval systems find and cite documents, how to evaluate a model honestly, and how all of this can be attacked.
The name comes from packet sniffing: capturing traffic to see what really crosses the wire. The same attitude applies to AI. Look at the tokens, the retrieved passages and the numbers, not just the final answer.
What you will find
- 17 articles that start with a direct answer and then explain the mechanism.
- 4 learning paths that order those articles into a sequence.
- 3 labs you run on your own computer with free tools.
- 8 browser tools that make ideas such as tokens and chunking visible.
- An AI tutor that answers from this material and shows its sources.
Editorial standards
- Every article answers a real question, and says so in its first lines.
- Claims about research name the work they come from.
- Code examples are run before they are published.
- Some drafts are written with help from AI tools. A person reviews every article before it is published, and is responsible for it.
- Articles show a publication date, and an updated date when the content changes.
- No invented statistics, testimonials or credentials.
How the tutor stays honest
The tutor only answers from passages it retrieves from this site. If nothing relevant is found, it says so instead of guessing, and it shows the passages it used for every answer. Read the details on the project page.
Security ethics
Security content here is written for defenders, students and authorised testers. Techniques are only to be used on systems you own or have explicit permission to test. See the disclaimer.