OpenAI is undergoing a strategic shift, discontinuing its Sora AI video generator, ending a $1 billion partnership with Disney, and scaling back ChatGPT's Instant Checkout feature. These changes aim to concentrate resources on core text and code models, potentially in preparation for a future IPO, and focus on the development of the upcoming 'Spud' model.
Sources: The Neuron, The Rundown AI
Google Research has developed TurboQuant, a compression algorithm that reduces AI model working memory by 6x without sacrificing accuracy. This innovation has the potential to significantly decrease inference costs for AI products and is being implemented in open-source models.
Sources: The Neuron, The Rundown AI
Apple is planning to integrate third-party AI assistants like Gemini and Claude into Siri through 'Extensions' in iOS 27, ending its exclusive partnership with OpenAI. This strategic move will position the iPhone as a platform for various AI models, offering users more choice in their AI assistant.
Sources: The Neuron, The Rundown AI
Google has launched Gemini 3.1 Flash Live, a new low-latency model designed for voice and vision agents, featuring improved function calling and extended conversation memory, available across Gemini Live, Search Live, and AI Studio. Google is also expanding its Search Live feature, powered by Gemini 3.1 Flash Live, to over 200 countries and territories.
Sources: DeepMind Blog, Google AI Blog, Latent Space, The Neuron
Elon Musk unveiled Terafab, a $25 billion facility designed to produce 50 times the current global AI compute capacity annually. The facility will integrate logic, memory, packaging, and testing for chips used in Tesla vehicles, Optimus robots, and space-grade AI satellites, potentially shifting compute to space.
Sources: The Rundown AI
Anthropic is reportedly considering an IPO as early as Q4 2026, with bankers estimating a valuation exceeding $60 billion. This potential offering would be one of the largest ever, indicating strong investor confidence in the generative AI sector.
Sources: The Neuron
Arm has introduced its first in-house chip, the 136-core AGI CPU, specifically designed for AI inference after 35 years of licensing designs. Meta is the launch customer, with OpenAI, Cerebras, and Cloudflare also signed up, marking Arm's direct entry into the AI hardware market.
Sources: The Neuron
OpenAI is partnering with Helion to secure 5GW of fusion power by 2030 and is also seeking to raise approximately $4 billion from private equity firms for enterprise joint ventures, offering 17.5% guaranteed returns. This strategy underscores OpenAI's long-term vision for sustainable, high-energy computing for AI and its aggressive fundraising efforts.
Sources: The Neuron
SoftBank has begun construction on a $500 billion AI data center campus in Ohio, aiming for 10 gigawatts of power. This significant investment demonstrates a major commitment to building the infrastructure needed for advanced AI development.
Sources: The Neuron
The US White House has released its initial national AI policy framework, advocating for federal preemption over state AI laws and stating that training AI on copyrighted material is legal. The framework aims to centralize AI regulation at the federal level and prevent a patchwork of state laws.
Sources: The Neuron, The Rundown AI
The NousResearch Hermes Agent autonomously wrote, typeset, edited, and published a 79,000-word novel titled 'The Second Son of the House of Bells.' This demonstrates a significant advancement in AI's creative and autonomous content generation capabilities, particularly in long-form narrative.
Sources: The Neuron
Research from the UK government's AI Security Institute shows that successive generations of frontier AI models are significantly outperforming predecessors in simulated multi-step cyberattacks. This indicates that AI systems are nearing autonomous, end-to-end cyberattack capabilities.
Sources: Import AI