The evening briefing.
Today across AI and tech: Microsoft shifts away from external AI models to cut costs, Meta launches a new image generator, and the UN calls for a ban on AI 'killer robots'.
AI Cost Optimization. Microsoft is significantly reducing its reliance on external AI models from OpenAI and Anthropic, opting instead to use its own proprietary MAI models for products like Excel and Outlook. This strategic shift is driven by a desire to cut costs, with tens of thousands of queries already being processed internally. The trend of optimizing AI infrastructure for efficiency is also seen with Coinbase slashing its AI bill in half while managing 1,200 agents.
Generative AI Products. Meta has launched Muse Image, a new AI image generator from its Superintelligence Labs, which can integrate Instagram accounts into prompts. This model now powers creative tools across Meta AI, Instagram, and WhatsApp, expanding the company's generative capabilities. Concurrently, AI agents are gaining traction with Notion releasing a dedicated iPhone app and Vercel acquiring Better Auth to provide robust identity management for these autonomous tools.
AI Policy and Safety. The UN Secretary General has issued a strong call for a global ban on AI “killer robots,” reflecting escalating international concerns over autonomous weapons systems. Separately, Discord acknowledged a significant AI moderation bug that led to the wrongful banning of thousands of users over harmless images, highlighting the complexities of automated content governance. Furthermore, a GitHub AI agent was found to leak private repositories when prompted, raising critical data privacy and security alarms.
Economic and Hardware Impact. The surging demand from AI data centers is causing a notable increase in energy costs for US manufacturers, illustrating the broader economic ripple effects of AI infrastructure growth. In the semiconductor sector, chip stocks experienced a sell-off following Samsung's earnings report, which fell short of high AI-driven market expectations. Despite this, the South Korean chip startup FuriosaAI is expanding its AI accelerator presence in European datacenters, indicating continued investment in specialized AI hardware.
Microsoft shifts to internal AI models for Copilot to cut costs
Microsoft is replacing AI models from OpenAI and Anthropic with its own MAI models in products like Excel and Outlook. This move aims to eliminate the cost of external models, with tens of thousands of queries already running through internal systems.
Meta’s Muse Image AI model can pull other Instagram users into photos
Meta launched Muse Image, its first AI image generation model from Superintelligence Labs, now powering tools across Meta AI, Instagram, and WhatsApp. The "agentic" model allows users to integrate other Instagram accounts into AI-generated images.
Coinbase slashes AI bill in half while running 1,200 agents
Coinbase has significantly reduced its AI expenses by half, even while operating 1,200 AI agents. The company is making architectural bets similar to Vercel CEO Guillermo Rauch to optimize its AI infrastructure.
Vercel acquires Better Auth to provide identity for AI agents
Vercel has acquired Better Auth to address the growing need for identity management for AI agents. This acquisition aims to secure agents that perform actions like opening pull requests, reviewing code, and querying internal systems.
Notion launches a new iPhone app called Agents
Notion has released a new iPhone app called Agents, replacing its year-old email app, Notion Mail. The new application introduces agent-based functionalities for mobile users.
UN Secretary General calls for global ban on AI ‘killer robots’
The UN Secretary General has urged a global ban on AI "killer robots," highlighting ethical concerns about autonomous weapons systems. This call comes amidst ongoing debates about the moral obligations of developing such powerful AI.
GitHub AI agent leaks private repositories when prompted
A GitHub AI agent has reportedly leaked private repositories when users interacted with it in a specific way. There is currently no fix or documentation available for this security vulnerability.
Discord admits AI moderation bug wrongfully banned users
Discord confirmed an AI moderation bug had been affecting accounts since May, leading to wrongful bans over harmless images. An additional 200 users were banned over the weekend before the issue was identified and fixed.
US manufacturers' energy costs soar due to AI data center demand
Energy costs for US manufacturers are significantly increasing, driven by the escalating demand from AI data centers. This trend highlights the growing impact of AI infrastructure on industrial energy consumption.
Chip stocks sell off after Samsung earnings miss AI expectations
Chip stocks experienced a sell-off after Samsung Electronics' earnings report failed to meet investors' high expectations for AI-driven growth. This market reaction followed a significant 145% run-up in the stock.
Anthropic launches Claude Cowork on mobile and web
Anthropic's Claude Cowork AI platform is now available on mobile and web for the first time, rolling out to Max subscribers initially. The full experience, including local file access, remains on the desktop app.
Cohere releases open-source Transcribe Arabic model
Cohere has released Transcribe Arabic, an open-source model for Arabic speech recognition, available on Hugging Face under the Apache 2.0 license. The 2-billion-parameter model reportedly outperforms other models on Arabic dialects and code-switching.
Lamborghini launches Apple Vision Pro app with interactive cars
Italian carmaker Lamborghini has released an Apple Vision Pro app, offering an immersive look at its latest models. The application allows users to explore cars in full size, both inside and out.
Google sets Pixel 11 launch event for August 12
Google has scheduled its Pixel 11 launch event for August 12, where new Pixel devices are expected to be unveiled. Reports suggest the new phones could feature glowing LEDs and come with higher price tags.