Bilingual

Meta AI chief says their coming LLM has caught up with OpenAI's flagship model
18 days ago
- Meta's AI chief Alexandr Wang claims the upcoming 'Watermelon' model has caught up with OpenAI's GPT-5.5 on benchmarks.
- Watermelon uses significantly more computational power than its predecessor Avocado (Muse Spark) and is currently in training.
- An update to Muse Spark is coming soon, with improvements in coding and agentic capabilities to rival competitors like Anthropic's Claude.
- Meta has aggressively invested in talent and infrastructure, offering top AI researchers large compensation and increasing infrastructure spending forecasts.
- Despite past struggles to match leaders, Meta aims to close the gap with OpenAI, Google, and Anthropic, with Zuckerberg driving the AI push.
The US-China AI arms race has taken an unexpected turn
7 days ago
- DeepSeek's open-source R1 model released in January 2025, which rivals top US AIs and is free, caused a $1 trillion drop in US tech stocks and prompted US lawmakers to propose banning it on government devices.
- Z.ai's GLM-5.2 release in contrast did not cause similar panic, indicating a shift in the US-China AI arms race where China now catches up, similar to other technologies like smartphones and electric cars.
- Chinese AI models are typically open-source and free for local use, while US models are often cloud-based, fee-charging, and closed, leading to differences in accessibility and innovation strategies.
- GLM-5.2 ranks as the most intelligent open-source AI by benchmarks, outperforming GPT-5.5 in software engineering tasks but slightly lagging in intelligence tests, highlighting industry-wide standard testing issues.
- DeepSeek's latest model leads in usage on OpenRouter, with seven of the top 10 LLMs built by Chinese companies, showing growing adoption despite potential security and compliance concerns for large enterprises.
- Experts warn that US restrictions on AI access may backfire by pushing others to develop independent ecosystems, as seen with Huawei, while China's open-source approach pressures Western closed models.
- Business inertia and risk aversion in big companies favor established US tech firms like Microsoft due to compliance and reliability, even if Chinese models are free and capable on laptops.
- Europe risks becoming an 'AI colony' dependent on foreign systems without developing its own AI industry under European legislation, facing a national security threat more serious than the nuclear arms race.
EU will force Google to share search data and open up AI on Android
5 days ago
- The European Commission has introduced new DMA measures requiring Google to enhance interoperability and competition on Android and search.
- Google must open Android to third-party AI assistants, allowing users more choice beyond its Gemini AI, which currently has preferential access.
- Google argues the changes could compromise privacy and security, but as a designated 'gatekeeper,' compliance with the legally binding measures is mandatory.
- The EU aims to prevent limitations on innovative AI services, citing that 60% of EU Android users are affected by current restrictions.
- The measures are designed to protect user privacy and device integrity, though concerns remain about data usage by generative AI models.
Chinese AI model Kimi K3 halts new signups amid skyrocketing demand
a day ago
- Moonshot AI paused new subscriptions to Kimi K3 due to overwhelming demand exceeding computing capacity.
- Kimi K3 is a 2.8 trillion-parameter open-weight model, noted as the largest of its kind and topping some benchmarks.
- High demand reflects intense AI rivalry between China and the US, with Chinese models gaining international traction.
- Experts suggest Moonshot underestimated popularity and compute demands, leading to infrastructure strain.
- The surge in competitive Chinese AI models pressures US tech stocks by threatening pricing power.
- Other Chinese releases like Alibaba's Qwen3.8 Max and Zhipu's GLM-5.2 also draw global attention.
Ben Thompson is wrong: US frontier labs are right to be panicking
21 hours ago
- Ben Thompson argues US frontier labs have lower unit costs and tool stickiness, but the author disagrees.
- Unit costs: US labs have higher electricity, data center costs, and may lose advantage as Chinese chips improve.
- Stickiness: AI tools like Claude Code are quickly replaced; users switch easily for better alternatives.
- US labs face structural issues: no secret sauce, high energy costs in a constrained grid, and overreliance on future profits.
- Predictions: US labs won't save by costs/stickiness, will lose frontier status, and China will lead global inference.
Gary Marcus: US won't "win" AI war with China. What we should do instead.
11 hours ago
- U.S. AI industry faces challenges as Chinese models like Kimi K3 match U.S. capabilities, being open-weight and free
- OpenAI and Anthropic's business models are questioned, with potential IPOs threatened by competition and lack of moat
- U.S. government's focus on LLMs and close ties to Silicon Valley led to strategic blunders, squandering lead
- Seven options proposed: do nothing, outlaw open source, build regulatory moat, bail out labs, ban Chinese models, buy out labs as national labs, or make AI a global public good
- International cooperation for AI as a public good, inspired by CERN and supported by China, is seen as most promising long-term solution

Hasty Briefsbeta

#ai competition

Meta AI chief says their coming LLM has caught up with OpenAI's flagship model

The US-China AI arms race has taken an unexpected turn

EU will force Google to share search data and open up AI on Android

Chinese AI model Kimi K3 halts new signups amid skyrocketing demand

Ben Thompson is wrong: US frontier labs are right to be panicking

Gary Marcus: US won't "win" AI war with China. What we should do instead.