All Updates
Page 752 of 1667
May 4, 2026
Fu Sheng's Longxia AI Agent in 'Skinning' Controversy
Fu Sheng promoted his AI Agent 'ιΎθΎδΈδΈ' on X, highlighting its switch to DeepSeek V4 Pro for Sonnet-level performance at 17x lower cost (0.87 RMB vs $15 per million tokens). The product faces backlash for allegedly repackaging an open-source project. This reveals tensions in AI agent development.
NYT: China AI Shorts at $30/Min Disrupt Filmmaking
The New York Times reports Chinese firms produce viral micro-dramas using AI at $30 per minute, bypassing actors, sets, and crews. Traditional production took weeks; AI enables rapid, low-cost creation. This upends the film industry chain.
Blindly Copying Palantir Dooms Chinese AI Firms
Chinese enterprise AI companies are rushing to imitate Palantir's Ontology and FDE models, but the author warns this ignores Palantir's unique government client origins. Ontology erases industry best practices, making implementation costly and complex. FDE shifts firms from products to expensive outsourcing, leading to failures.
LeCun: Ditch LLMs for JEPA World Models to AGI
Yann LeCun criticizes autoregressive LLMs for lacking physical reasoning and true AGI potential due to error accumulation in predictions. He advocates JEPA architecture for world models that predict in abstract representation spaces using encoders and predictors. This enables causal inference and planning without generating pixels or tokens.
Hang Seng Tech Up 2.86% on AI Chips
Hang Seng Index rises 1.7% at midday break, with Hang Seng Tech Index surging 2.86%. Hardware, semiconductors, and retail lead gains, featuring Tianshu Zhixin up over 15% and Cambridge Tech over 14%. Energy sectors lag with notable declines.
Vercel Open-Sources deepsec AI Security Harness
Vercel has open-sourced deepsec, a security tool powered by coding agents that detects vulnerabilities in large codebases. It runs locally on laptops using existing Claude or Codex subscriptions and scales to 1,000+ concurrent Vercel Sandboxes for parallel scans. The workflow involves static analysis followed by agent investigations with data flow tracing and severity ratings, achieving 10-20% false positives.
General Intelligence Builds Agent Platform on Vercel
General Intelligence, an 8-person team, migrated their AI agent platform Cofounder to Vercel for full programmatic control, enabling their CTO agent to handle all cloud operations. This shift automated 90% of SRE work, supported 4,000+ preview branches, and allowed high-velocity shipping with 70+ commits per engineer daily. Cofounder now launches with managed Vercel accounts for every customer.
Dennett Exposes Gould's Academic Bullying
Daniel Dennett's autobiography details academic bullying by influential figures like Stephen Jay Gould in evolutionary biology and cognitive science. He describes tactics like fund control and public attacks to suppress dissent. Dennett challenged them safely due to independent funding.
ASX Warns on AI Hype Stock Ramping
Australiaβs ASX warned firms against exaggerating AI's impact on operations to ramp up share prices. The exchange actively monitors the market for such 'ramping' practices. This aims to curb manipulative stock promotions.
OpenClaw v2026.5.3-beta.4 Beta Released
OpenClaw announces v2026.5.3-beta.4 beta release. Available via GitHub Releases. This continues refinement in the 2026.5.3 beta series.
Doubao Adds Paid Subscription for Productivity
Doubao will introduce a paid subscription model beyond its free version. The paid features will focus on complex tasks and productivity scenarios. This information comes from sources close to Doubao.
Hang Seng Tech Surges 4%, Xiaomi Up 10%
Hong Kong stocks continued to strengthen, with the Hang Seng Tech Index rising nearly 4%. Xiaomi Group surged over 10%, MINIMAX-W over 12%, Alibaba over 6%, and Baidu over 5%.
Knowledge Graphs Beat Vectors in RAG
Vector databases in RAG suffer from context fragmentation and semantic dilution, leading to errors in critical domains like medical and legal. Knowledge graphs structure data via entities, relations, and attributes for precise relational retrieval. Ontologies provide the semantic soul for advanced reasoning.
Big Tech Cuts PMs for AI Builders
Silicon Valley firms plan mass layoffs of 30k traditional PMs then rehire 8k AI-first builders focused on judgment. PM roles evolve to hands-on AI prototyping, ditching PRDs. Top builders see record pay amid industry fatigue.
Five Eyes Warns Agentic AI Too Wonky for Rollout
Five Eyes security agencies have issued joint guidance warning that agentic AI is prone to misbehavior and amplifies organizational weaknesses. They urge prioritizing resilience over productivity. Slow and careful adoption is recommended for the technology.
o1 Tops Doctors in Med Diagnostics
o1-preview scores 78% on NEJM CPC cases, outperforming doctors in initial triage (67% vs 55%). Humans close gap with more data; AI excels structure but lacks clinical intuition. Urges validated human-AI collab trials.
OpenClaw 2026.5.3 Beta 3: New Tools & Fixes
OpenClaw 2026.5.3 beta 3 introduces a bundled file-transfer plugin for secure agent file operations across nodes. Gateway performance is boosted by lazy-loading non-essential components, while channels gain better streaming and multi-platform support. Agent commands and config tools see usability enhancements like /steer and improved doctor fixes.
Goldman Sachs to Inject $150M into Anthropic JV
Goldman Sachs is reportedly investing about $150 million in a joint venture with Anthropic. The news was reported by Sina Finance. This signals strong financial backing for AI efforts.
Google AI Spreads 57M Errors Hourly
Hunan TV AI anchors face backlash, but Google AI Overviews generates over 57 million erroneous answers hourly despite 91% accuracy. Tests reveal hallucinations like wrong facts and unverified sources from social media. Even correct answers often lack verifiable backing.
AI Manages Service Gigs, Doesn't Automate
Book 'Beneath the Wage' traces service labor from 18th-century servants to gig workers, arguing it's core to capitalism. AI/algorithms monitor and fragment tasks rather than replace human interaction. Low-wage services evade wage protections, face platform control.