AI Daily Briefing — 2026-07-17

Andrew Dai, a former researcher at Google DeepMind, has achieved a remarkable feat in the AI startup world. Just months after leaving the tech giant, Dai raised a $55 million seed round for his new company, Elorian, at a staggering $300 million valuation—all before releasing a single product. The fundraise, which involved strategic investors like Nvidia and Menlo Ventures, signals intense investor appetite for cutting-edge AI ventures.

Dai believes visual AI represents the next major frontier in artificial intelligence, an area he says has seen uneven progress compared to math, physics, and coding. "At Elorian, we want to build models that will advance us toward visual AGI," Dai explained in an interview. His decade of experience at DeepMind, including work that later influenced ChatGPT, helped him articulate a complex technical vision into a compelling narrative for investors.

Rather than chasing the highest valuation, Dai prioritized partners who understood the realities of building frontier AI. He emphasized speed as a critical competitive advantage and shared strategies for recruiting top researchers from Big Tech. The episode offers practical lessons for founders navigating today's fast-moving AI landscape, including how to pitch technical products to nontechnical investors and build durable moats as technology evolves.

While the AI industry races to brand its latest breakthroughs as 'AGI' or 'superintelligence,' Alexandre LeBrun, CEO of AMI Labs, firmly avoids both terms. In an interview with TechCrunch AI, LeBrun dismissed the labels as meaningless marketing shifts. 'There's no good definition. What is superintelligence? I don't know. It's not a very useful word,' he said, noting that the industry simply swaps one vague term for another.

LeBrun's company, founded by Yann LeCun, focuses on world models—AI systems that predict physical outcomes rather than just text. Unlike large language models (LLMs) that guess the next word, world models anticipate real-world changes, like a glass tipping off a table. LeBrun sees this as critical for robotics, where current machines run 'completely static' routines and remain 'really dumb in the physical world.' He cited a recent incident where a dancing robot kicked a child, arguing that context-aware AI could have prevented it.

LeBrun insists world models and LLMs are complementary, not rivals. LLMs excel at language, while world models provide real-world understanding. He believes almost any industry 'that touches the real world' could benefit, from manufacturing to healthcare. Drawing on his experience at health AI startup Nabla, he compared today's AI to a doctor trained only on textbooks—useful but missing the hands-on residency needed for true competence.

AMI Labs remains pre-product but is already scouting partners in robotics, manufacturing, and electronics. LeBrun emphasized that world models cannot be built in isolation; they require real environments to train on. 'Robots are not safe right now,' he warned. 'There's no solution for that today.'

Apple’s generative AI platform, Apple Intelligence, has secured regulatory approval for launch in China, marking a significant milestone for the tech giant in one of its most important markets. The Cyberspace Administration of China gave the nod following a deal that integrates Alibaba’s Qwen AI model into Apple’s operating systems, including iOS, iPadOS, macOS, and visionOS. Baidu also confirmed it is collaborating with Apple to develop AI features tailored for Chinese users, according to a spokesperson.

The partnerships, rumored for months, come as Apple works to adapt its AI offerings to meet local regulations. The company had faced delays in rolling out Apple Intelligence in China due to a lack of regulatory clearance. Alibaba stated that its Qwen models will bring capabilities like text and image understanding and generation to Apple devices, though no specific timeline was provided. Apple is also reportedly exploring integrations with DeepSeek and ByteDance.

The approval arrives at a pivotal time for Apple in China. The company generated $20.5 billion in sales in Greater China during the second quarter, a 28% year-over-year increase, and recently reclaimed the No. 2 spot in the country’s smartphone market after promotional discounts boosted iPhone sales. The AI launch is expected to further strengthen Apple’s competitive edge in the region.

Both Alibaba and Baidu confirmed their respective deals, with Alibaba noting that its models will be embedded into Apple Intelligence experiences. The collaborations highlight Apple’s strategy of partnering with local AI leaders to navigate China’s strict internet regulations while expanding its AI footprint in a key market.

OpenAI has developed a new artificial intelligence system named GPT-Red, designed to act as a super-powered hacker. Its sole purpose is to attack other AI models from the company, probing for weaknesses and security flaws. This automated process, known as red-teaming, was previously a task handled by human experts. By pitting AI against AI, OpenAI aims to strengthen its models against real-world cyber threats, according to an exclusive report from MIT Technology Review AI.

Meanwhile, heat pumps are experiencing a surge in popularity across the United States, even as a key federal tax credit for the appliances has expired. Sales of these highly efficient electric heating and cooling systems have doubled over the past 15 years. In the first quarter of 2026, heat pumps outsold traditional natural-gas furnaces by a significant 32%, signaling a major shift in home energy use despite the removal of financial incentives.

This trend underscores a growing consumer preference for efficient electric technology over fossil fuels. The continued rise of heat pumps suggests that market forces and long-term energy savings are now driving adoption more than temporary tax breaks. The technology's ability to both heat and cool homes efficiently makes it an increasingly attractive option for homeowners looking to reduce their carbon footprint and energy bills.

These two stories highlight contrasting facets of modern technology: one involves using advanced AI to secure other AI systems, while the other shows a clean energy technology gaining mainstream traction. Both developments point to a future where automation and efficiency play central roles, whether in cybersecurity or home climate control.

PsiQuantum, a startup founded in 2016 by four UK physicists, aims to build a revolutionary quantum computer using photons instead of traditional qubits. The machine would be housed in a facility resembling a data center mixed with an ice cream factory, featuring around 100 stainless-steel cabinets kept near absolute zero with liquid helium. Inside, hundreds of chips would guide thousands of light particles through optical switches and beam splitters, with precise measurements enabling calculations that would take classical computers millions of years.

Quantum computers leverage the ability of quantum bits to exist in multiple states simultaneously, unlike classical bits limited to 0 or 1. While current prototypes remain too small and error-prone for practical use, PsiQuantum claims its approach could dramatically accelerate tasks like drug development. For instance, predicting how cytochrome P450 enzymes break down medications—a process that currently takes over a decade—could be reduced to just four minutes, according to Philipp Ernst, the company's vice president of quantum applications.

PsiQuantum has attracted over $1 billion in funding and partnerships with major chip manufacturers, using existing semiconductor fabs to build its systems. The company broke ground on a Chicago facility last year and plans a second site in Australia, operational by 2027. It is one of only two firms, alongside Microsoft, to reach the third stage of a rigorous government evaluation program. However, verifying progress remains challenging, as quantum computing advances are incremental and opaque. The coming year may reveal whether PsiQuantum's years of secretive work will yield a commercially useful machine or fall short.