All Updates
Page 296 of 921
April 7, 2026
Sakana AI's AI Detects SNS Misinformation
Sakana AI announced a new AI technology to counter misinformation on social media networks. It centers on three key functions: information visualization, truthfulness judgment, and diffusion suppression simulation. This development stems from a project commissioned by Japan's Ministry of Internal Affairs and Communications.
Claude Research Omits Chinese Citation, Apologizes
Claude's viral research on AI emotion circuits failed to cite a prior Chinese team's contributions. Facing backlash, Anthropic issued a formal apology. This marks the first systematic study of AI emotional feedback loops.
Google AI Overviews Accuracy Questioned
Google's AI Overviews deliver authoritative-looking answers but draw from a wide range of sources, including trustworthy sites and unreliable Facebook posts. This sourcing approach raises significant concerns about the factual accuracy of the generated summaries.
Intel Goes All-In on AI Chip Packaging
Intel is fully committing to advanced chip packaging technologies. The company seeks to capitalize on the booming AI market for growth. This strategic shift positions Intel to compete in high-performance computing for AI.
Force-Motion Control Fusion Evolves Robot Brains
Explores differences between force control and motion control in embodied intelligence. Discusses their fusion as the breakthrough for evolving the 'small brain' in robots. Seeks ultimate solutions for advanced robotic control.
Is ACL Dominated by Benchmarks Now?
Reddit discussion questions if premier NLP venue ACL has shifted to benchmark-focused papers. Post-results feeds are filled with benchmark titles. Notes young researchers submitting 10+ papers per author.
Gemma 4 Had Hidden MTP Removed for Compatibility
Gemma 4 models include MTP weights for speculative decoding in LiteRT files, but Google removed them intentionally for broader compatibility. A Google employee confirmed this after errors on Pixel 9. Community calls for full model release or reverse engineering.
Token Surge Reshuffles China Internet Rankings
The token wave is rewriting China's internet underlying logic. The decade-old valuation framework is loosening amid AI-driven shifts. Internet giants face reordered standings.
Trivy supply chain attack breaches EU Commission data
EU Commission cloud environment breached via supply chain attack on open-source Trivy tool. TeamPCP hackers used tainted Trivy to steal AWS API keys. They extracted 92GB compressed data (340GB uncompressed) from Europa.eu infrastructure.
AI Coding Surge Sparks Bug Crisis
AI tools have democratized coding, enabling more people to generate code. However, this leads to increased bugs, vulnerabilities, and security risks. There's now a shortage of engineers to review the influx of AI-produced code.
26% Gen Z Dating AI for Emotional Ties
A survey reveals 26% of Gen Z are dating AI, finding conversations easier than with real people. These interactions provide emotional and romantic fulfillment beyond sex. The trend underscores broader loneliness in society.
Horizon's First Full Year: ~10B Revenue, 30.8% AD Margin
Horizon Robotics, led by founder Yin Qi, reports nearly 10 billion RMB revenue in its first full year with the Journey team. Intelligent driving gross margin reached 30.8%. This marks the first public disclosure of smart driving income.
Qwen Code v0.14.1-preview.0: Fixes & CLI Enhancements
Qwen Code releases preview v0.14.1 with bug fixes across CLI, hooks, subagents, Telegram, and auth. New features include follow-up suggestions in CLI/webui, enhanced /btw side questions, and mid-turn queue drain for agents. Additional improvements cover rate-limit skipping via Ctrl+Y and tool refactoring.
AI Shoulder Pet for GenZ Socializing
Blinkko is an AI magnetic shoulder pet acting as a social signal for young users, using lightweight AI for voice emotion detection and responses. App leverages LLMs for matching owners and events. Incubated by Qiji Chuatan, targets North America launch with hardware-subscription model.
MoE Models Converge on 10B Active Parameters
Recent MoE models like Qwen 3.5 122B and MiniMax M2.7 230B activate around 10B parameters despite varying total sizes. Training cost scales with active params and tokens, explaining economic convergence at ~9e23 FLOPs. Questions inference memory scaling with expert count.
Why AI Glasses Over Phones & PCs?
XREAL founder Xu Chi is interviewed on why AI glasses can replace smartphones and computers. The piece is from Ifanr's 'Diversity Company' series. It challenges traditional device usage paradigms.
Pixel Cake 9.0: AI Revolutionizes Photo Retouching
Pixel Cake 9.0 heralds a disruptive revolution in AI-powered photo retouching. Retouchers are shifting from operators to managers and from technical executors to creative decision-makers.
Goldman Sees Tech Valuation Opportunities
Goldman Sachs strategists highlight attractive valuations in tech due to hyperscale data center capex concerns. Tech P/E ratios trail peers like consumer and industrial sectors despite robust growth. Sector not in bubble; geopolitics enhances appeal via resilient cash flows.
Windows 11 Introduces Feature Flags
Windows 11 will add feature flags for experimental updates. Users can enable them without waiting for gradual rollouts. The change sparks cautious optimism.
ModelBest Secures Funding, Hits Unicorn Status
Chinese on-device AI foundation model developer ModelBest raised several hundred million RMB (tens of millions USD) in funding. The round was led by Shenzhen Capital Group and Inovance Capital. This marks its entry into unicorn status after three funding rounds in the past year.