THE DIFF

US Federal Right to Privacy Act Drafted

Draft legislation for a US federal privacy law has been published on GitHub. The ‘Right to Privacy Act’ proposes explicit rights for individuals over their data, including access, correction, deletion, and portability. It aims to establish a national standard, overriding the current patchwork of state laws like CCPA/CPRA, and would impose strict consent and data minimization requirements for services. This draft is a significant, if overdue, step towards a US equivalent of GDPR.

Source: Hacker News


Canada Bill C-22 Enables Warrantless Metadata Access

Canada’s newly proposed Bill C-22 includes provisions that compel telecommunication service providers to store customer metadata and provide it to law enforcement without a warrant. Critics argue this creates a backdoor for mass surveillance by requiring providers to build systems capable of intercepting and retaining vast amounts of user data on demand. The bill has significant architectural implications for any service operating with Canadian users.

Source: Hacker News


RARE EARTH

The Push Against ‘Sloppypasta’

A new site, stopsloppypasta.ai, serves as a manifesto against the low-effort, verbose, and often incorrect code generated by LLMs from lazy prompts. It advocates for treating AI as a tool that requires skilled operation, not a magic box. The site provides concrete examples of ‘sloppy’ vs. ‘structured’ prompts, encouraging engineers to provide more context, constraints, and examples to get deterministic, high-quality output. It’s a welcome push for craftsmanship in an era of automated boilerplate.

Source: Hacker News


TOOL OF THE WEEK

A CLI-First LLM Workflow

This is a practical, no-nonsense workflow for integrating LLMs directly into a command-line development loop. The author uses llm-cli to pipe code and context directly to models, combined with a custom Python script that acts as a pre-processor. The script automatically includes relevant file contents, a tree structure of the current directory, and a persistent ‘system prompt’ to maintain context. It’s a powerful approach for treating the LLM as a filter in a Unix-style pipeline, not a chat partner.

Source: Hacker News


MAG 7 NEWS

Google Signals The End of ‘AI Wrappers’

A Google and Accel India accelerator reviewed over 4,000 AI startup applications, ultimately selecting just five. Organizers noted that approximately 70% of the pitches were thin ‘wrappers’ around existing large language models. The selected companies focus on foundational models or highly specialized applications, signaling a clear preference from major investors for deep tech over simple API integrations. This is a market correction you can build a career on.

Source: TechCrunch AI