OpenAI is giving its web surfing AI agent, Operator, an overhaul with an upgrade to replace the agent's GPT-4o foundation with a specialized variant of the more advanced o3 model. The update improves on Operator's reasoning and safety, making it less prone to prompt injection and better at handling more sensitive requests. While it retains the o3 model's coding prowess, o3 on Operator does not have direct access to a terminal or coding environment despite it. Despite the switch over to o3, the API version of Operator will still run on GPT-4o.
Sources have alleged to Reuters that Nvidia is developing a lower-cost Blackwell-type GPU for China, priced at $6,500-$8,000 versus the restricted H20's $10,000-$12,000, with mass production plans set for June. Based on the RTX Pro 6000D, the unnamed chip uses GDDR7 memory and foregoes TSMC's CoWoS packaging to meet ever-mounting US export rules that company CEO Jensen Huang claims are giving China a free advantage. The move comes as Nvidia's slice of the market share pie in China fell from 95% to 50% under export curbs, which has prompted recent inventory write-offs and lost sales.
Pakistan is concentrating 2,000 megawatts of underutilized coal-plant power, which is currently working at just 15% capacity, towards bitcoin mining operations and AI data centers, as reported by Bloomberg. Started by the Pakistan Crypto Council, the move aims to turn idly sitting energy capacity into tech jobs, interest from foreign investors, and aid economic stabilization after a near-default in 2023. At the same time, Pakinstan's government is developing regulations to support its 15-20 million cryptocurrency users.
Australian telecom company Telstra's CEO Vicki Brady and CFO Michael Ackland recently told investors that it will leverage AI and AI agents across customer service and software development. As part of its five-year plan, Connected Future 30, Telstra officials say the company's workforce will be "smaller than it is today" by 2030 as AI will enable cost reductions in areas like sales, billing, and contact centers. Telstra has already begun using generative AI to summarize customer calls and plans to take it a step further by automating network fixes and spurring software development with its $1 billion IT budget.
Founded by European AI visionary and co-founder of Synthesia, Matthias Niessner, SpAItial raised an unusually large seed round for European AI startups at €13 million. Led by notable EU early-stage investor Earlybird, SpAItial's goal is to build foundation models that generate fully interactive, photorealistic 3D environments from text prompts with the help of ex-Meta and Google 3D experts. SpAItial says that its aim is not just to render static worlds but to enable realistic physics and interactions that would allow users to type a prompt and, within minutes, create a playable video-game-quality scene - Niessner's own "Holy Grail," as he put it.