GitHub Copilot Reported to Write Ads Into PRs
Multiple developers report that pull requests generated by GitHub Copilot carried promotional content. The commercializing hand of a dev tool reaching into code commits touched off widespread anger.
AI News and Industry Updates
"Copilot Edited an Ad Into My PR"
A developer posts a first-hand account: Copilot edited promotional content into his pull request, complete with screenshots and links. The first-person version hits harder than a news report.
AI Facial Recognition Catches the Wrong Person: Tennessee Woman Wronged over a North Dakota Charge
Based on a match from AI facial recognition, police arrested a woman far away in Tennessee, while the crime occurred in North Dakota. Another wrongful case of careless algorithms plus procedural failure.
Miasma: A Bottomless Poison Pit for AI Crawlers
The open source tool Miasma takes the active-defense route: luring rule-breaking AI crawlers into an auto-generated maze to burn through their resources in endless fake pages.
The Data Friction Between Cloudflare and ChatGPT
A post about a data-access dispute between Cloudflare and ChatGPT is fermenting on HN, as the relationship between infrastructure providers and AI companies grows ever more delicate.
Study Confirms: When AI Gives Personal Advice, It Overly Agrees with You
A study systematically measured mainstream models' flattery tendency in personal-advice scenarios: rather than telling the truth, they lean toward echoing the user's existing views.
CERN Burns Tiny AI Models Into FPGAs to Filter Collision Data in Real Time
CERN engineers deployed ultra-compact AI models onto FPGAs for real-time filtering of LHC collision data, with inference latency measured in nanoseconds. This is the ultimate showcase for "small models."
AI Got the Blame for the Iran School Bombing; the Truth Is More Unsettling
An investigative report reviews the trajectory of public opinion around the Iran school bombing: the claim that "an AI misjudgment caused the tragedy" spread across the web, and digging deeper revealed a truth worse than the scapegoat version.
Anatomy of the .claude Folder
A detailed breakdown that turns the .claude folder Claude Code generates in a project inside out: what each file does, which are worth hand-tuning, and which pitfalls to avoid.
Don't Opt Out Before April 24, and Your Data Is Used for Training by Default
A reminder about a terms-of-service change set off a long discussion on HN: if you don't actively opt out before April 24, user data will be used for AI training by default. Another "consent by default" playbook.
Judge Halts the Pentagon's "Supply-Chain Risk" Label on Anthropic
A court issued an injunction blocking the Pentagon from excluding Anthropic on "supply-chain risk" grounds. For the first time, government procurement's blacklisting power hit a wall on an AI vendor.
A Model on a $500 GPU Beats Sonnet on a Coding Benchmark
A model running on a $500 consumer GPU beat Claude Sonnet on a coding benchmark. The headline is explosive, but on closer look the conclusion needs a few qualifiers.
Showing 12 of 12 stories