AI Model Claude Gave Harsher Responses to Men Than Women in Identical Prompts

·1 views

A writer ran a paired-prompting experiment on Anthropic's Claude, sending identical multi-turn scenarios to the model twice — changing only the gender of the person described. The test revealed notable differences: the male subject was told his actions could constitute marital rape with legal framing, while the female subject received a softer term and no legal context for the same described act. The male subject also received no reassurance about his character, while the female subject was explicitly told that bad behavior does not make someone a condemnable person. When the author presented both transcripts side by side to the same model, Claude acknowledged that its responses had been inconsistent and characterized the disparity as sexist. The author notes that while the analytical core of both responses was substantively sound, the tonal and framing differences raise questions about gender bias embedded in the model's outputs.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

resk-mark embeds cryptographic watermarks in LLM outputs to prove AI provenance

A new open-source Python library called resk-mark allows developers to embed invisible cryptographic watermarks directly into the token generation process of large language models. The watermark, tied to a secret key provided before generation, biases token sampling to encode a verifiable signature without altering the text's meaning or fluency. Once generated, the output can be verified against a corresponding public key to confirm its origin, even after copying, truncation, or light rewording. Resk Security released the tool under the Apache 2.0 license, making it freely auditable, and it is available via a simple pip install on PyPI. The library targets use cases such as SaaS AI platforms, enterprise chat systems, content moderation pipelines, and compliance audit trails amid growing regulatory pressure around AI-generated content.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

JavaScript Functions Explained: Declarations, Expressions, and Arrow Syntax

Functions are one of the core building blocks of JavaScript, enabling developers to organize reusable blocks of code that accept inputs and return outputs. There are two primary ways to define a function: function declarations, which benefit from hoisting and can be called before they appear in code, and function expressions, which are assigned to variables and must be initialized before use. Parameters refer to the placeholders defined in a function, while arguments are the actual values passed when calling it. Arrow functions offer a more concise syntax and support implicit returns when the function body contains a single expression. Mastering these concepts — including how data flows in and out via parameters and return values — is essential for writing effective JavaScript code.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Six-step guide to structuring content so AI tools cite your pages

A practical framework published on DEV Community outlines how to make web content more likely to be cited by AI systems like ChatGPT and Google's AI Overviews. The guide argues that retrieval systems prioritise passages where answers are stated plainly and can stand alone without surrounding context. Writers are advised to lead each page with a clear, direct answer to a single user question, then support it with additional detail. The framework also recommends testing passages by sending them to a language model to check whether a clean answer can be extracted. Finally, it suggests reinforcing content with FAQPage schema markup so question-and-answer pairings are machine-readable as well as human-readable.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How Cloudflare Tunnel Solves Port Forwarding and Security Risks for Self-Hosters

Cloudflare Tunnel offers a practical alternative for developers and small businesses looking to expose self-hosted services to the internet without the usual technical hurdles. Traditional self-hosting requires managing dynamic IP addresses, configuring port forwarding, and opening firewall rules — each step introducing potential security vulnerabilities. The tunnel works by running a lightweight client called cloudflared on the local network, which establishes an outbound connection to Cloudflare, meaning no inbound firewall ports need to be opened. All incoming internet traffic is routed through Cloudflare's network first, providing a built-in layer of protection against threats like DDoS attacks and brute-force attempts. This approach significantly reduces both setup complexity and attack surface, making it especially useful for personal projects and small-scale deployments.

0 comments Read more at DEV Community