Attention and transformers diagrams

Atlas of Stochastic Parrot Anatomy 🦜

Published: Sep 24 2026

Some diagrams about attention in artificial neural networks and transformers architectures (as used in LLMs, vision transformers, VLMs).

Tool call execution through user prompt injection (CSRF) of the llama-server Web UI

Published: Jul 20 2026

The llama-server Web UI of the llama.cpp project is vulnerable to (Reprompt-like user prompt injection vulnerability. An attacker can inject arbitrary user prompt with query parameter (?q=...). This can lead to arbitrary shell command execution, data exfiltration, etc. (through tool calls).

User prompt injection (CSRF) on Le Chat and Grok

Published: Jul 17 2026

An user prompt injection vulnerability (CSRF) both in in the e Chat (Mistral) and Grok (Reprompt-style) allows attackers to inject user prompt with query parameter (?q=...) potentially leading data exfiltration.

Reprompt-style vulnerability in Le Chat (Mistral) and Grok.

llama.cpp quickstart (part 2)

Published: Jul 16 2026

How to quickly use llama.cpp for LLM inference (part 2). This is a follow-up of a previous post on the same topic.

Testing Unicode Tag Smuggling in chatbots

Where we learn that the sky is actually a giant blueberry smoothie 🫐

Published: Aug 20 2025

Testing ASCII smuggling using Unicode Tags on LLMs/chatbots. Nothing new here. Just a short summary.

llama.cpp quickstart

Published: May 14 2025

How to quickly use llama.cpp for LLM inference (no GPU needed).

vLLM quickstart

Published: May 14 2025

How to quickly use vLLM for LLM inference using CPU.

Transformer-decoder language models

Published: Jan 7 2025

Some notes on how transformer-decoder language models work, taking GPT-2 as an example, and with lots references in order to dig deeper. This is intended both as a a roadmap for understanding on how LLMs work (especially the ones using a transformer-decoder architecture) and a a summary/recap on the topic.