This week, AI accelerated its super intelligence ambitions by metapoaching two of the top minds of the meta, but Mistral launched its first open audio AI model aimed at enterprise users. Chinese robotics company Aimoga is expanding its humanoid retail robots worldwide, AWS has announced plans for a human-equipped AI agent market, and Nvidia is preparing a new export-controlled AI chip for China.
Research warns that Openai, Deepmind, and humanity that monitoring the inner reasoning of AI can soon become impossible without urgent action.
Meta recruits two major Openai researchers to accelerate their close efforts
According to Wired, Meta has been hired by Openai researchers Jason Wei and Hyung to join Chung for the new Superintelligence Lab. Both play an important role in the development of Openai's O-series models and are considered strategic adoption to promote meta into artificial general information. (AI Insider)
Mistral launches Voxtral, the first open audio AI model for business
French startup Mistral has launched Voxtral, a family of open-weight audio models designed for affordable, advanced voice AI tasks such as transcription and running voice commands. With Mistral reportedly seeking $1 billion in new funding, it is available by embracing the Face and Mistral platforms, aiming to challenge closed AI systems. (AI Insider)
Items from Chinese Robotics Company introduce the development of car dealers humanoid robots
Chinese robotics company AiMaga is expanding globally with humanoid robots, and is now used by car dealers to promote sales and facilitate staff workloads. In partnership with Chery Auto Group, Aimoga aims to expand CheryGPT-powered robots into retail, service and housing markets, as it will cover a 50 billion euro industry by 2035. (AI Insider)
AWS launches AI Agent Marketplace as an important partner for humanity
Amazon Web Services launched the AI Agent Marketplace on July 15th, allowing startups to distribute autonomous AI agents directly to AWS Enterprise customers. AWS aims to partner with humanity to streamline the distribution of agents and expand the adoption of agent-based APIs. (AI Insider)
Nvidia's new AI chips target the Chinese market under export restrictions
Nvidia plans to release a unique Chinese AI chip in September based on the Blackwell RTX Pro 6000, which was modified to comply with US export controls as early as September, according to the Financial Times. The chip omits advanced features such as NVLink and high-bandwidth memory, reflecting Nvidia's strategy of being present in the Chinese data center market despite regulatory limitations. (AI Insider)
Research and innovation
AI's inner monologue provides new clues for safety, but the windows may be closed
New research from Openai, Deepmind, and Anthropic found that monitoring inferences in AI systems' mindsets can help detect inconsistencies and harmful behaviors. Researchers warn that this transparency could fade as models evolve and maintain active work for AI safety and encourage active work. (AI Insider)
Policy and Governance
Top AI researchers are calling for an industry-wide focus on mentality monitoring to improve AI safety
AI researchers such as Openai, Deepmind and Anthropic are collaboratively calling for emergency research into thinking (COT) monitoring to improve transparency in next-generation AI inference models. Their position paper warns that without collaboration, advances in model design could undermine internal decision-making, safety and alignment efforts (AI Insider)
Startups and Capital
MaintInX raises $150 million to transform asset management and industrial operations with AI
Maintenance raised $150 million in Series D funding and reached a $2.5 billion valuation to advance AI-powered asset management and machinery health platforms. Serving more than 11,000 businesses, the system uses predictive AI to reduce downtime and optimize industrial maintenance operations around the world. (AI Insider)
Thinking Machine Lab closes $2 billion funding round at $12 billion valuation to build safer and scalable AI systems
Founded by former Openai executive Mira Murati, Thinking Machines Lab raised $2 billion at a $12 billion valuation just months after its launch, despite not having the product yet. Supported by investors such as Andreessen Horowitz and Nvidia, the startup aims to develop safer, general purpose AI systems with an open source focus. (AI Insider)
OpenEvidence secures a $210 million Series B at a $3.5 billion valuation and expands its AI healthcare platform for physicians
OpenEvidence, a medical search engine for AI-powered doctors, has raised $210 million in Series B funding with support from Google Ventures and Kleiner Perkins, with a $3.5 billion valuation. Used by more than 40% of US clinicians, the platform has recently launched DeepConsult, an AI agent that provides evidence-based responses and provides summary of advanced medical research. (AI Insider)
Rwazi announces $10 million funding round to replace all intestinal calls with AI copilot
Rwazi has raised $12 million to expand its AI Copilot and Simulation platform. This provides real-time market insights from direct consumer behavior data. Used by Fortune 100 companies, this system helps businesses detect shifts, simulate results, and optimize decision-making across marketing, product and operations. (AI Insider)
