The morning briefing.
While you slept: Ford's AI automation backfires, Anthropic's Fable 5 may return, and China's Lingsheng supercomputer tops global charts.
AI Governance. The Trump administration is reportedly preparing to lift restrictions on Anthropic's Fable 5 model, potentially making it available again within days after safety concerns led to its temporary halt. This highlights ongoing tensions between rapid AI development and regulatory oversight. Meanwhile, a Chinese cybersecurity firm is developing AI tools to rival Anthropic's Mythos, framing the competition as "cyber-nuclear deterrence." This underscores the escalating geopolitical dimension of AI.
AI Performance & Efficiency. New research continues to push the boundaries of AI capabilities and efficiency. Peking University and DeepSeek open-sourced DSpark, a speculative decoding framework that significantly boosts LLM inference speed by up to 85%. Separately, Sina's VibeThinker-3B, a small 3-billion parameter model, shows logical reasoning compresses well, matching much larger models on benchmarks. However, a Princeton University study found most AI models fail a 500-day startup survival test, indicating struggles with complex business operations.
AI Infrastructure & Supply Chain. The immense demands of AI are reshaping global hardware and supply chains. China's new Lingsheng supercomputer now tops global charts with 219 EFLOPS, showcasing a significant leap in compute power. Optical module MCUs are critical as AI data centers drive demand for 800G and 1.6T connections. Chinese chipmakers are also investing in silicon carbide (SiC) semiconductors to improve data center energy efficiency. Apple is reportedly seeking permission to buy memory from a blacklisted Chinese supplier.
Economic & Societal Impact. The broader economic and societal impacts of AI are becoming increasingly apparent. The Bank for International Settlements (BIS) warns that the AI investment boom, channeled through loosely regulated non-bank channels, could lead to a sharper financial crash during a downturn. In a notable incident, Ford's attempt to replace human workers with AI backfired badly, leading to significant operational issues. Tech firms are also blaming AI development for rising device and console prices, passing costs onto consumers.
Ford's AI Automation Backfires After Sacking Human Workers
Ford hired AI and sacked humans, but the automation backfired badly. This led to significant operational issues for the company.
Anthropic's Fable 5 Could Return as Trump Administration Prepares to Lift Restrictions
Anthropic's AI model, Fable 5, could be available again within days, as the Trump administration is close to lifting restrictions imposed on June 12 over safety concerns. The Pentagon and NSA still need to sign off on the decision.
Apple Vision Pro Executive Reportedly Joins OpenAI's Hardware Team
Paul Meade, Apple's vice president in charge of the Vision Pro headset, is reportedly leaving the company to join OpenAI’s hardware team. This move signifies a notable talent shift between major tech players.
Apple Seeks Permission to Buy Memory from Blacklisted Chinese Supplier CXMT
Apple is seeking an exception from the Trump administration to buy RAM chips from CXMT, a Chinese company blacklisted by the Pentagon. This move aims to alleviate pressure on its supply chain amid skyrocketing memory prices.
Peking University and DeepSeek Open-Source DSpark for LLM Inference Efficiency
Peking University and DeepSeek jointly open-sourced DSpark, a speculative decoding framework that boosts LLM inference speed by 60-85%. It also delivers up to 661% throughput gain under strict latency constraints.
Sina's VibeThinker-3B Shows Reasoning Compresses Well in Small LLMs
Sina Weibo's VibeThinker-3B, with just three billion parameters, matches much larger models like DeepSeek V3.2 on math and coding benchmarks. Researchers propose that logical reasoning compresses well into small models, but broad world knowledge does not.
Only Three AI Models Survived a 500-Day Startup Test Above Starting Capital
Researchers at Princeton University developed CEO-Bench, a test where AI agents run fictional software companies for 500 simulated days. Most current models went broke, and a simple rule-based heuristic outperformed nearly all of them.
Chinese Cybersecurity Firm Builds AI Tools to Rival Anthropic's Mythos
360 founder Zhou Hongyi presented two AI security tools designed to compete with Anthropic's Mythos, with one already flagging 3,432 vulnerabilities. Zhou compared Mythos to "cyber nuclear weapons" and called for China to build its own strategic deterrent.
China's Lingsheng Supercomputer Claims World's Top Spot at 219 EFLOPS
China's domestically developed Lingsheng supercomputer has claimed the world's top spot with 219 EFLOPS. This marks the country's return to the No.1 ranking since Sunway TaihuLight in 2017.
Optical Module MCUs Critical as AI Data Centers Drive 800G and 1.6T Demand
MCU chips for optical modules are emerging as a critical semiconductor segment due to the AI data center buildout. This is driving a surge in demand for 800G and 1.6T connections.
Chinese Chipmakers Bet on Silicon Carbide for AI Data Center Energy Efficiency
As AI puts intense pressure on data center energy grids, some Chinese chipmakers are betting on highly efficient silicon carbide (SiC) semiconductors. Shenzhen-based Basic Semiconductor is pursuing an IPO in Hong Kong to fund its expansion.
BIS Warns AI Boom Exposes Investors to Risk of Sharp Crash
The Bank for International Settlements (BIS) warned that the AI investment boom, with capital flowing through loosely regulated non-bank channels, could lead to a sharper and faster crash than a traditional banking crisis during a downturn. This was stated in its annual economic report.
Tech Firms Blame AI for Rising Device and Console Prices
Xbox consoles, Nintendo's new Switch 2, and Valve's Steam Deck are among the gadgets seeing price hikes in recent months. Tech firms are attributing these increases to the demands of AI development.
Tesla Settles Lawsuit Over Fatal Pedestrian Crash Involving Full Self-Driving
Tesla has settled a lawsuit stemming from a deadly incident in Arizona in 2023 involving its Full Self-Driving technology. Details of the settlement were not disclosed.