Dan Goodin writes that ASCII smuggling, a technique once primarily used to hide malicious prompt injections from Large Language Models (LLMs), has been adopted by spammers to evade email filters. By using invisible Unicode tags that mimic the structure of standard text, attackers can bypass machine learning-based spam detectors and natural language processing models without alerting human readers. This method allows words like "funding" or "credit" to be broken into non-standard tokens that escape keyword detection while remaining perfectly readable to a person once rendered in an email client.
- ASCII smuggling uses the Unicode Tags block, which contains 128 characters designed to be invisible to humans but readable by computers.
- Microsoft observed spam signatures using this technique spike from 21,000 per day to over 2.5 million within a four-day period in early February.
- The method is effective against modern AI-driven filters because it disrupts the way tokenizers process words into sub-word pieces.
- While originally used for stealthy prompt injections, spammers now use it specifically to obfuscate financial keywords from automated detectors.
- ASCII smuggling uses the Unicode Tags block, which contains 128 characters designed to be invisible to humans but readable by computers.
- Microsoft observed spam signatures using this technique spike from 21,000 per day to over 2.5 million within a four-day period in early February.
- The method is effective against modern AI-driven filters because it disrupts the way tokenizers process words into sub-word pieces.
- While originally used for stealthy prompt injections, spammers now use it specifically to obfuscate financial keywords from automated detectors.
>"Using DSPy to automatically create, evaluate, and optimize your prompts"
Manual prompt engineering is often slow and unreliable due to unpredictable inputs. DSPy addresses this by treating prompt development like traditional ML training. It automatically generates, evaluates (using "LLM-as-a-judge"), and optimizes prompts based on high-level task descriptions, providing a faster, more systematic way to build robust LLM applications.
Manual prompt engineering is often slow and unreliable due to unpredictable inputs. DSPy addresses this by treating prompt development like traditional ML training. It automatically generates, evaluates (using "LLM-as-a-judge"), and optimizes prompts based on high-level task descriptions, providing a faster, more systematic way to build robust LLM applications.
Learn how to label text without the need for task-specific training data by using zero-shot text classification. This guide explains how pretrained transformer models, such as BART, reframe classification as a reasoning task where labels are treated as natural language statements.
Key topics include:
* The core concept of zero-shot classification and its advantages for rapid prototyping.
* Using the Hugging Face transformers pipeline with the facebook/bart-large-mnli model.
* Implementing multi-label classification for texts belonging to multiple categories.
* Improving accuracy through custom hypothesis template tuning and clear label wording.
Key topics include:
* The core concept of zero-shot classification and its advantages for rapid prototyping.
* Using the Hugging Face transformers pipeline with the facebook/bart-large-mnli model.
* Implementing multi-label classification for texts belonging to multiple categories.
* Improving accuracy through custom hypothesis template tuning and clear label wording.
This document defines the "cid:" and "mid:" URL schemes for referencing messages and their parts. The "cid:" scheme is used within multipart messages, like emails, to reference embedded content such as images. The "mid:" scheme allows referencing entire messages or specific parts within them using their Message-ID and Content-ID. It details the syntax for these URLs, emphasizing the importance of unique Content-IDs and the handling of encoded characters. The specification requires implementations to support the full "mid" URL form (message-id/content-id) for interoperability.
This article details how to set up an email triage system using Home Assistant and a local Large Language Model (LLM) to summarize and categorize incoming emails, reducing inbox clutter and improving email management. It covers the setup of a REST command to interface with Ollama, the automation process, and the benefits of using a local LLM for privacy.
A Model Context Protocol (MCP) server that provides tools for interacting with JMAP (JSON Meta Application Protocol) email servers. Built with Deno and using the jmap-jam client library.
DispatchMail is an open source locally run (though currently using OpenAI for queries) AI-powered email assistant that helps you manage your inbox. It monitors your email, processes it with an AI agent based on your prompts, and provides a (locally run) web interface for managing drafts/responses, and instructions.
This page details the topic namers available in Turftopic, allowing automated assignment of human-readable names to topics. It covers Large Language Models (local and OpenAI), N-gram patterns, and provides API references for the `TopicNamer`, `LLMTopicNamer`, `OpenAITopicNamer`, and `NgramTopicNamer` classes.
This tutorial demonstrates how to perform semantic clustering of user messages using Large Language Models (LLMs) by prompting them to analyze publicly available Discord messages. It covers methods for data extraction, sentiment scoring, KNN clustering, and visualization, emphasizing faster and less effort-intensive processes compared to traditional data science approaches.
The article discusses how AI agents can help business owners manage their inboxes by sorting emails, filtering spam, and automating responses. It highlights the benefits of AI in maintaining productivity through deep work, updating business processes, and enabling personalized communication. The author suggests implementing safeguards to ensure accuracy and appropriateness in email communications.
