Briefings

The morning briefing.

While you slept: Chinese LLMs aim for

RIGHT NOW, IN ONE BREATH

Chinese AI Advancement. The global race for AI supremacy continues, with a notable debate between Elon Musk and Zhipu AI CEO Tang Jie regarding the timeline for Chinese large language models (LLMs) to reach "fable level" capabilities. This discussion follows Zhipu AI's release of GLM-5.2, which is seen as a significant step. Chinese developers are actively accelerating their efforts in trillion-parameter foundation models to narrow the gap with leading US rivals amidst increasing export controls.

AI Safety Research. Efforts to enhance AI safety are progressing, with OpenAI researchers demonstrating that targeted "beneficial trait" training can make AI models broadly safer and more resistant to manipulation. This approach, which differs from Anthropic's constitution-based method, improved deception detection and overall performance across numerous benchmarks. Concurrently, new research explores how LLMs can detect their own epistemic blind spots, aiming to reduce uncertainty in structured tasks like clinical data analysis.

Chip Supply Constraints. The surging demand for AI infrastructure is placing significant strain on the hardware supply chain, leading to tangible impacts on consumer electronics. Nothing's co-founder announced the cancellation of the CMF Phone 3 Pro, citing the ongoing RAM crisis exacerbated by AI's compute needs. This comes as global optical chip makers, including NVIDIA with a $2 billion investment, accelerate capacity expansion to meet the demands of AI data centers, while Europe's silicon photonics firms warn of inadequate infrastructure to commercialize research.

Digital Platform Oversight. Regulatory bodies are increasing scrutiny on digital platforms and AI systems, with developments in the UK and India. The UK's top data and AI regulator has resigned following "inappropriate" humor, highlighting the sensitivity of governance roles. Google is appealing a German court ruling that held it directly liable for inaccurate AI-generated search results. Separately, Telegram faces questions from Ofcom over illegal incitement, and the Delhi High Court upheld a decision to restrict the app in India.

64
Sources scanned
5,311
Headlines processed
#172
Edition
15.6k
Discussing now

RAM crisis kills Nothing’s next budget phone, CMF Phone 3 Pro scrapped

The ongoing RAM crisis, fueled by AI demand, has led to the cancellation of Nothing's CMF Phone 3 Pro. Co-founder Carl Pei stated they could not build a phone that felt like a genuine step forward.

Musk and Zhipu AI CEO debate Chinese LLM progress to "fable level"

Elon Musk and Zhipu AI CEO Tang Jie publicly debated the timeline for Chinese large language models to reach "fable level" capabilities. This discussion follows Zhipu AI's release of GLM-5.2.

OpenAI researchers show "beneficial trait" training makes AI models broadly safer

OpenAI researchers demonstrated that reinforcement learning on desired behavioral traits like truthfulness and corrigibility improves AI model safety. This method works across domains and enhances deception detection.

UK's top data and AI regulator quits after 'inappropriate' humor

The head of the UK's data watchdog, who has been in the role since January 2022, resigned after admitting to making inappropriate attempts at humor. This marks a significant leadership change in AI regulation.

Google appeals ruling making it directly liable for AI-generated search overview content

Google is appealing a German court ruling that held it directly liable for inaccurate AI search results. The AI had falsely linked two Munich-based publishers to fraud schemes.

US and ASML dispute top chip tool presence in China

The US government claims ASML's top chip tool may be in China, a claim ASML denies. This dispute highlights ongoing tensions over semiconductor export controls.

Delhivery launches AI-native mapping platform for logistics businesses

Logistics giant Delhivery announced the launch of ‘Delhivery Maps’, an AI-native suite of geospatial APIs. This platform is designed for commercial logistics and navigation.

Quantum Systems and Tencore expand defence robotics production with new German hub

Quantum Tencore Industries, a German-Ukrainian joint venture, will deliver 2,000 TerMIT unmanned ground vehicles to the Ukrainian Armed Forces. This expansion includes a new German manufacturing hub.

Festina Finance secures over €25 million to modernise pensions and life insurance platforms

Copenhagen-based FinTech startup Festina Finance secured over €25 million in growth investment from Birchway Capital. The funding will modernize cloud-native modular technology platforms for pensions and life insurance.

Barret Zoph is out at OpenAI again after just five months

Barret Zoph, OpenAI's head of enterprise AI sales, has departed the company after only five months. He had returned to OpenAI in January after co-founding Thinking Machines Lab.

Telegram questioned by Ofcom after arsonist recruited on app

Ofcom has questioned Telegram on its methods for detecting illegal incitement after an arsonist was directed on the app by a handler linked to Russia. This follows a conviction for attacks on properties associated with Keir Starmer.

Global optical chip makers accelerate capacity expansion for AI data centers

Coherent, Nokia, JX Advanced Metals, and other optical chip leaders are expanding capacity due to surging AI data center demand. NVIDIA is investing $2 billion to secure photonic interconnect supply.

Chinese satellite engine costs plummet, boosting commercial space sector

China's commercial space industry is achieving new milestones with dramatic cost reductions in domestically developed satellite propulsion systems. This accelerates the sector's growth.

Norway greenlights first full-scale ship tunnel

Norway has approved the construction of the world's first full-scale ship tunnel, designed to allow larger vessels to bypass dangerous waters. This project marks a significant engineering feat.