Hacker NewsTuesday, July 21

China's open AI strategy is outperforming America's closed approach. Chinese companies are releasing powerful AI models openly, turning US compute restrictions into a distribution advantage while commoditizing where American firms profit. Open technologies typically win infrastructure battles due to their permissionless nature and innovation potential. As Chinese models now match US frontier capabilities at lower costs, America's AI dominance is eroding. The locked-down US business model lacks real competitive moats and ignores ecosystem benefits, creating a losing strategy that could damage the broader US economy if AI spending collapses.

The content only contains the title "Airport Simulator" without any additional meaningful information to summarize.

A mathematician claims to have disproven the Jacobian Conjecture by providing a specific polynomial map from C³ to C³ with Jacobian determinant -2 that sends three distinct points to the same point, contradicting the conjecture's requirements for invertible maps.

A hacker breached Romania's cadastre agency and wiped the entire land registry database after a failed extortion attempt, bringing the country's real estate market to a standstill. The attacker, identified as ByteToBreach from Algeria, used valid credentials to access systems and delete data and backups. Other notable incidents include Hugging Face being hacked by an AI agent, Coca-Cola suspending production after ransomware hit its Fairlife subsidiary, and various government policy changes including the UK scrapping its digital ID scheme and France banning Polymarket.

The European Commission is negotiating an Enhanced Border Security Partnership with the US that would grant American authorities access to EU biometric databases to maintain visa-free travel. EDRi warns the leaked draft shows the EU caving to excessive US demands for traveler surveillance and profiling, potentially violating fundamental rights and EU law. The agreement could enable discrimination based on political opinions and social media activity, with the US threatening to end visa exemptions as leverage.

Xiaomi-Robotics-1 is a robot foundation model that scales policy models by combining large-scale embodiment-free pre-training with real robot data. Using 100,000 hours of diverse trajectory data across 1,700 scenarios, it demonstrates clean scaling behavior where larger models and more data lead to better real-world robot performance, achieving state-of-the-art results on simulation benchmarks and efficient adaptation to new tasks.

Researchers used an AI prompt adapted from OpenAI's mathematical problem-solving approach to discover a critical WordPress vulnerability affecting over 500 million instances worldwide. The AI found a pre-authentication SQL injection flaw in WordPress's batch API that can be escalated to remote code execution, demonstrating how advanced AI can identify zero-day vulnerabilities in widely-used software.

This is a critical security review of OpenCode, a popular open-source AI coding agent with 161k GitHub stars. The author tested it with a local LLM and found it to be fundamentally flawed with severe security vulnerabilities and poor design decisions that make it unsafe to use.

Users report significant problems with OpenAI's automatic context compaction feature, which loses important details and can't be disabled. Many prefer Anthropic's larger context windows despite performance degradation at higher token counts.

Researchers from French and Italian universities found that AI advice dramatically reduced people's willingness to admit ignorance, dropping from 44% to just 3%, while accuracy fell from 27% to 9%. Despite being wrong more often, participants' confidence increased from 30% to 76% when using AI assistance.