Mysterious shape on Mars in new NASA picture; debate over alien civilization intensifies

Google co-founder Sergey Brin has in recent months strongly urged key AI employees to focus entirely on the company's Gemini model as it strives to catch up with competitors, according to two people familiar with his remarks.

When Anthropic was advancing with a preview of its powerful Claude Mythos model, Brin addressed hundreds of employees at a company-wide meeting in April, calling on Google DeepMind's AI lab to work faster, one anonymous source reported.

Google's Gemini model briefly surpassed competitors last November, but new updates from Anthropic and OpenAI have once again placed it in a catching-up position. According to one source and another person knowledgeable about the situation, Google postponed the release of a new flagship Gemini version by two months in August after internal testing showed that its performance continued to lag behind rivals in areas such as coding.

Restructuring Leadership at Google DeepMind

On August 5, the Mountain View, California tech giant announced a major reorganization of the leadership of its AI division, Google DeepMind. As part of this restructuring, leader Demis Hassabis stepped down in favor of his deputy, Karan Kavukchuoğlu. Hassabis took the role of chairman, taking on fewer management responsibilities. Simultaneously, two original Gemini technical leads left to found a startup.

Although some industry observers question whether Google is falling out of the model race, interviews with seven people familiar with Gemini development efforts indicate that Google's commercialization strategy and the recent staff reorganization are linked to its drive for model superiority.

These changes mark a further erosion of DeepMind's autonomy, which Google has gradually weakened since acquiring the London lab in 2014 as part of a continuous effort to use its AI across most of its software and increase profits. The restructuring included information shared at the general meeting on August 6 that some teams are moving from DeepMind to corporate Google, according to two sources.

Commercial Interests

Google declined to answer questions regarding Brin's recent involvement, the structural changes in DeepMind, and employee concerns.

Under Hassabis's leadership, Google DeepMind undertook a broader range of AI research projects while simultaneously creating Gemini models vital to corporate commercial interests. However, in previous years, Hassabis opposed some initiatives that could have brought new revenue to Google or helped strengthen its position in the AI race. His allies at Google previously claimed that he was not neglecting business or focusing too much on long-term science.

In 2025, Google CEO Sundar Pichai appointed Kavukchuoğlu, who was then DeepMind's technical director, to an additional senior role as Google's Chief AI Architect. His task was to integrate Gemini into all company products. Four sources reported that in the following months, the influence of some leaders at DeepMind diminished, while Kavukchuoğlu became the central figure determining Gemini's direction.

One source noted that Kavukchuoğlu had Brin's support. The news of the restructuring caused Google's stock to drop by 4% on the day of the announcement.

At the company-wide meeting last Thursday, Hassabis and Kavukchuoğlu attempted to ease employee concerns—most of whom learned about the change when it became public—about the future of the AI lab. Google leaders stated that daily operations in the lab would remain the same, according to three sources with direct knowledge of the meeting. Sources reported that Kavukchuoğlu would remain in charge of Gemini; he took over oversight of the model's development more than a year ago, according to five sources.

Hassabis delivered a presentation on how AI can advance science and change society—long-term issues that will be the focus of his new role as chairman, according to three sources familiar with the meeting. Despite this message, new concerns arose at the general meeting that DeepMind's autonomy is being absorbed by Google. Hassabis and Kavukchuoğlu informed employees that several non-technical teams are being moved from DeepMind to the corporate reporting structure, according to two sources. For an organization that once encouraged employees to pursue auxiliary projects, the message of regaining dominance in models was also evident.

Leaders, including Pichai, have stated in recent months that Google considers itself at the forefront in certain areas, such as video generation, and remains committed to catching up in other fields despite setbacks.

Meanwhile, senior leaders in the cloud division expect Kavukchuoğlu's appointment to alleviate tension within DeepMind regarding the allocation of Google's limited supply of AI chips, known as Tensor Processing Units (TPUs), another source reported.

Boundaries of Progress

Google's restructuring occurs against the backdrop of Brin's efforts to mobilize personnel, something he also did after OpenAI launched ChatGPT in 2022, which shook Google.

Although he no longer holds an executive position, the 52-year-old Brin unofficially influences model training efforts. Over several months, he has told key personnel that they need to reach cutting-edge milestones, according to one source familiar with his efforts. Brin uses the hidden power he possesses as a Google co-founder to guide resource allocation in specific areas, such as recursive self-improvement—a point where technology can improve without human intervention, this source reported. Google declined to provide Brin for an interview.

Meanwhile, Hassabis reduced his involvement in managing Gemini and handed over this responsibility to his deputies shortly after initially launching the AI model in 2023, according to two sources.

Although Google caught up to the cutting edge with the release of Gemini 3 last November, it quickly fell behind again. Four sources attributed its lag to persistent structural issues that initially led to it. For example, computational power limitations and disagreements among many Gemini project leaders resulted in fewer resources for critical areas like coding, according to two sources. Google's bureaucratic nature also led to it releasing models slower than more agile competitors, according to four sources.

Google declined to comment on these claims, but Pichai told Wall Street analysts last month that the company will continue to build out its infrastructure to remove constraints around TPUs, which it needs for training models, as well as for consumer and enterprise products.

Against this backdrop, Kavukchuoğlu is consolidating his power. He moved from London to Mountain View last year upon promotion to Chief AI Architect, reporting to Pichai. Some leaders from London found their influence on Gemini's technical direction diminished after Hassabis reduced his involvement in Gemini, and Kavukchuoğlu moved to Google headquarters, according to a source with direct knowledge.

Kavukchuoğlu has also begun building a reputation in other parts of Google as a person more focused on commercialization. Even before his move, Kavukchuoğlu, not Hassabis, was the main contact between DeepMind and Google Cloud, the company's primary AI revenue source, according to a source with direct knowledge. He will now make final decisions on major matters for DeepMind, a Google representative stated last week.

Some of the most prominent leaders, besides Kavukchuoğlu, including original Gemini technical co-leads Jeff Dean and Oriol Vinyals, left the company during last week's reorganization. Employees who reported to some of these leaders had a rare degree of autonomy to conduct research unrelated directly to Gemini training, according to three sources. After their departure, Google did not immediately share plans regarding the fate of these teams or projects, causing concern among some about the further erosion of the lab's bastion against commercial interests.

India utilizes its indigenous satellite navigation system, NavIC, developed by ISRO. This system covers the entire country and the surrounding region, ensuring India's positioning independence from systems controlled by other parties. However, the issue with using this system has always related to ground equipment, not aviation.

A satellite navigation signal is useless without a corresponding chip to receive it. These chips are produced by only a small number of companies, and there are almost none in India. Consequently, devices sold in the country could receive GPS signals but not NavIC, as commercial receivers supporting NavIC did not exist.

MumbaiSemi was founded in 2024 to address this commercial gap. Its founders spent years working on this problem at IIT Bombay.

Vijay Kanchetla received a Ph.D. in RF integrated circuit design at this institute and previously worked at Intel. Professor Rajesh Zele teaches in the Electrical Engineering department and holds a Ph.D. from Carnegie Mellon University. The company is located in the SINE incubator, the institute's technology business incubator.

The technology initially existed as a research project called Dhruva, funded by the Ministry of Electronics and Information Technology with SAMEER participating as the authorized agency. This project was created by students and researchers in Zele's laboratory. The first silicon sample was obtained from a fab in December 2019. Zele noted that the chip was designed in 18 months and became the first receiving chip from his lab fully developed by students.

The essence of the development lies in its reconfigurability. A standard receiver is built for specific frequency bands, so supporting multiple navigation systems usually requires multiple chips or hardware changes. The team's work describes a receiver that is fully programmable for the bands used by NavIC, GPS, Galileo, and BeiDou, without the need for external components or hardware changes on the chip between them.

The main engineering challenge relates to signal weakness. The signal coming from the satellite is significantly weaker than the surrounding noise. According to Zele's description of the chip, it is capable of separating the navigation signal and amplifying it by approximately 400,000 times before digital conversion on the same chip.

The commercial version is called DhruvaPro. MumbaiSemi claims that it has passed silicon validation, meaning it has been fabricated and tested, not just simulated, and the company holds two patents on this design. The smaller number of external components around the chip also leads to a more compact and less expensive receiver module, which is what the device manufacturer actually acquires.

The work received recognition within the institute: IIT Bombay awarded Zele and Kanchetla the Dr. P. K. Patwardhan Technology Development Award in 2024 for DhruvaPro during National Engineers Day. The research was also peer-reviewed. The paper describing the receiver, with Kanchetla as lead author and Zele among co-authors, was published in IEEE Transactions on Microwave Theory and Techniques in March 2022, which is a stronger form of external validation than most companies in this field can demonstrate.

The company participated in the Bharat Innovates exhibition in Nice in June 2026, where it sought partners in communications and navigation, rather than announcing clients. Nevertheless, the commercial stance remains closed: no data has been disclosed regarding clients, orders, revenue, or reliable funding amounts, and the company still needs to transition from silicon validation to mass production.

In situations where a soldier needs to know the situation beyond the next ridge, sending a patrol is slow and puts people at risk. Large drones are easily detected and audible from a distance, which negates the goal of covert observation. In such situations, a machine is needed that is small enough to be carried in a backpack and quiet enough not to attract attention.

These devices are created by the defense division of BotLab Dynamics, which is based at IIT Delhi. This company is known for using a thousand drones over Rashtrapati Bhavan during the Beating Retreat ceremony in 2022, as well as for record light shows.

Both parts of the business are based on similar core work. BotLab was founded in 2016 by Tamnaem Bunker, Anud Barnwal, and Dr. Sarita Akhawat. The swarm control software, which coordinates hundreds of drones into a formation over a stadium, is closely related to what is required to coordinate them over rough terrain.

In 2019, the company accepted a challenge from the Indian Air Force in Pokhran. According to its own data, the park that worked perfectly in Delhi failed in the desert because low-grade components used then perished from heat and dust. The company could not fulfill the task before senior military officials.

In response, the company began manufacturing its own parts. It now produces its own flight controllers, GPS modules, and electronic speed regulators instead of buying ready-made components. This distinguishes the machine operating in a demonstration from one capable of functioning in the field. Furthermore, it allows it to claim that its drones are manufactured in India, not assembled from imported parts.

The company's main defense product is Atri, a nano-class drone for reconnaissance and surveillance. It was developed with support from the iDEX grant awarded in 2023 to create India's smallest reconnaissance drone. The company describes Atri as designed for precise reconnaissance in high-risk areas and claims it can be quickly deployed to survey routes that forces need to monitor.

Alongside it is Vayusim, a training simulator that the company developed for defense sector users. It includes terrain and weather modeling, allowing operators to practice without the need for flights. Vayudh states that the company's systems have registered over 300,000 hours of autonomous flight, covering commercial drone shows, industrial inspection, and defense applications.

Vayudh raised $10 million, equivalent to about 83.5 crore rupees, in a round announced in April 2025, led by Dharana Capital. Participants in the deal included Dipinder Goyal, founder of Eternal Ltd, the parent company of Zomato, as well as Ajay Aggarwal from Action TESA, Sumit Jalan from LetsVenture, and Navam Capital. Previously, the company had received funding from Florintree Advisors.

Vamsi Duvvuri from Dharana Capital, commenting on the investment, pointed to India's rising defense spending and the push for domestic manufacturing as an opening market for the company, calling its swarm management technology field-tested. The funds will be directed towards research and development, increasing production capacity, and integrating systems into the armed forces. The company has about 200 employees, and its previous defense work included demonstrating heavy drones for lifting in Tvanga for the Army Design Bureau and showcasing swarm management systems to senior defense officials.

Atri was demonstrated to Prime Minister Narendra Modi and French President Emmanuel Macron at VivaTech in 2026. The company did not disclose the order volume, delivery volume, or client name, so the commercial aspect of this cooperation remains confidential.