GDSC Hult Weekly Update: April 7th - 14th Newsletter

‘Social Order Could Collapse’ in AI Era, Two Top Japan Companies Say

Wall Street Journal (April 7, 2024)

In a significant move, Japanese telecommunications giant Nippon Telegraph and Telephone (NTT) and media conglomerate Yomiuri Shimbun Group Holdings have jointly unveiled an AI manifesto. This manifesto highlights concerns about the uncontrolled proliferation of AI models and their potential threats to social cohesion and democratic principles. While acknowledging the productivity benefits of generative AI, the document also raises apprehensions about the AI initiatives of major US corporations.

The manifesto calls for immediate action to protect elections and national security from AI-related risks, aligning with global attention on AI legislation. Initiatives like the EU's ban on emotional recognition in educational and professional settings and President Biden's proposals to mandate AI companies to report potentially risky projects underscore the urgency. However, policymakers face challenges in balancing AI regulation with preserving fundamental liberties like free speech on social media.

NTT, with significant government ownership, and Yomiuri Shimbun Group Holdings, have been instrumental in shaping policy agendas and studying the implications of generative AI since 2023. While expressing concerns about AI's potential for disseminating harmful content and simulating fictitious political dialogues, they also recognize its transformative potential. NTT, for instance, utilizes AI technologies like its "tsuzumi" large-language model, akin to OpenAI's ChatGPT, to enhance efficiency while conserving computational resources, especially in Japanese-language contexts.

AI data security startup Cyera confirms $300M raise at a $1.4B valuation

TechCrunch (April 9, 2024)

Cyera, an Israeli-origin company, utilizes AI to help businesses localize and monitor their stored data within networks. Industry leaders like retired Sequoia executive Doug Leon endorse Cyera's vision and the importance of data management. Recently, Cyera secured $300 million in Series C funding, valuing the company at $1.4 billion, with investors including Sequoia, Accel, Redpoint, and Cyberstarts. Coatue led the round, with Spark Capital, Georgian, and AT&T Ventures also participating.

Cyera plays a critical role in safeguarding companies' data security, spanning from thwarting cyber attacks to preventing data leaks into extensive language models. The magnitude of these challenges within large enterprises is daunting to grasp. For instance, in 2019, AT&T suffered a data breach that exposed the details of 72.6 million current and former account holders, potentially leading to far-reaching consequences.

Cyera plays a crucial role in safeguarding companies' data security, countering cyber attacks and preventing data leaks. The platform's comprehensive evaluation of data origin, storage, and usage provides a solution for enterprises, especially amid the transition to online or hybrid work post-COVID-19. With data volumes predicted to soar, effective AI-driven posture management is essential to mitigate risks.

Drawing from their experience in the Israeli military and leveraging cutting-edge technology, the company's co-founders have successfully translated their insights into a thriving business, infusing entrepreneurial expertise into their endeavors. With nearly triple the funding from their previous $100 million Series B round in June 2023, Cyera aims to further enhance its offerings for clients, empowering them with greater control over their data.

Meta debuts new generation of AI chip

Reuters (April 10, 2024)

Meta has just unveiled more details about its internally developed artificial intelligence accelerator chips, internally codenamed 'Artemis', indicating a deep shift in its energy consumption strategy. These chips are poised to revolutionize Meta's reliance on energy-intensive NVIDIA AI chips, currently powering Facebook, WhatsApp, and Instagram's algorithms. Described by Meta as a blend of compute power, memory bandwidth, and capacity tailored for recommendation models and search ranking algorithms, Artemis promises groundbreaking efficiency.

The Meta Training and Inference Accelerator (MTIA) has focused on crafting software to leverage the new infrastructure efficiently. Simultaneously, Meta has been procuring chips from NVIDIA and other manufacturers. Taiwan Semiconductor Manufacturing Co. has committed to producing the Artemis chip using its advanced "5nm" process. Expected to triple the performance of its predecessor, this processor has already been deployed in Meta's data centers, with several initiatives underway. Meta is eager to accelerate the rollout of its projects, anticipating the transformative impact of these cutting-edge advancements.

A Chip Fab Business Grows in Brooklyn

Wall Street Journal (April 11, 2024)

Nanotronics, a groundbreaking startup founded in 2010 by Matthew Putman and his father in Ohio, is pioneering the integration of AI into various industries, fundamentally reshaping how we approach processes and structures. With a primary focus on chip manufacturing, the company's innovative strides are centered around its 'Cubefabs' located in Brooklyn, New York, strategically positioned adjacent to the Brooklyn Navy Yard.

These Cubefabs represent a departure from the traditional model of expansive manufacturing facilities, offering a more compact and modular approach. Backed by a substantial $162 million investment from prominent investors like Peter Thiel’s Founders Fund and Intel co-founder Gordon Moore, Nanotronics is poised to revolutionize production dynamics.

Already, the company is gearing up to deliver its first Cubefab unit to a client in the Middle East by 2025, signaling its rapid progression towards market implementation. Additionally, with the potential for support from the $53 billion CHIPS Act, Nanotronics stands to benefit from government initiatives aimed at bolstering national production capabilities.

One of the key features of Cubefabs is their advanced AI-powered software, which enables efficient operation with minimal defects. Moreover, this technology allows for the utilization of innovative materials like gallium oxide alongside traditional silicon carbide, promising chips with superior energy efficiency and diverse applications in sectors such as electric vehicles and power stations. Clients should pay a fee ranging from $400,000 to $3 million for hardware and AI-based software usage after the construction.

However, the implementation of Cubefabs comes with its own set of challenges. Nanotronics guides customers in selecting optimal locations for the facilities, considering factors such as water and electricity supply( 800 gallons of water a day and 2.7 megawatts of electricity), as well as navigating local regulations regarding chemical usage.

Energy-Guzzling AI Is Also the Future of Energy Savings

Wall Street Journal (April 12, 2024)

While the dark side of AI's coin has been laid bare with its voracious energy appetite, particularly evident in the consumption of data centers, such as the revelation that a single ChatGPT request consumes nearly 10 times more electricity than a standard Google search, as reported by the International Energy Agency. However, there's a glimmer of hope in the potential of AI optimization to mitigate these drawbacks. By harnessing AI's capabilities in optimizing household and commercial building energy usage, we can tip the seesaw back into balance, alleviating the adverse impacts of its energy consumption.

Despite the forecast of AI data centers consuming a significant portion of America's power by 2030, there's an opportunity for AI to tackle energy inefficiency elsewhere. In the real estate sector, which accounts for 30% of global energy consumption, a third of this energy is wasted due to inefficient systems, with a projected 1% annual growth. In 2023, $244 billion was spent trying to address these issues.

The pandemic-induced shift to remote work, evidenced by a 50% drop in US office occupancy, underscores the need for smarter energy management solutions, as buildings continue to consume energy despite reduced occupancy. Energy cost surges in Europe and tightening climate regulations in the US further emphasize the importance of energy efficiency. AI-driven solutions promise to revolutionize sustainable real estate management.

By leveraging data from existing smart devices, AI can analyze energy usage, predict future demands, and suggest optimizations. Real estate companies are already integrating these systems into new buildings, while retrofitting existing buildings with better software proves valuable. For instance, Schneider Electric helped Aspiria, an innovation campus in Kansas, reduce its energy bill by 16%, saving $1.5 million. This trend has boosted shares in power plant companies and AI-enabled energy efficiency firms, as owners expect these systems to enhance efficiency.

Vana plans to let users rent out their Reddit data to train AI

TechCrunch (April 13, 2024)

Vana, a startup founded by former MIT Media Lab students Anna Kazlauskas and Art Abal, aims to empower data creators and owners by offering them the opportunity to get rewarded the data they generate, which usually not happens when sold to big tech companies for model training. The platform, allows users to pool their data, including chats, speech recordings, and photos, into datasets for AI training. This data can then be accessed by app owners seeking to personalize user experiences, with reduced onboarding and computational costs.

Users have the option to self-host their data instead of storing it on Vana's servers, giving them greater control over its use by apps and developers. Vana charges a monthly subscription fee starting from $3.99 and a fee for app developers transferring datasets for model training, which is a reason to believe that the company has no incentive to exploit user data, according to Kazlauskas.

While Vana is not directly selling data, it has begun developing tools to enable users to do so, such as Reddit Data DAO. This program pools users' Reddit data, including karma and post history, allowing them to collectively decide how the data is used and be rewarded with cryptocurrency according to their Reddit’s karma. Although Reddit initially withheld data access for AI model training until its IPO, it now does so.

Nevertheless, Reddit has banned Vana's discussion hub about DAO and accused the startup of exploiting its data export system. However, Reddit emphasizes its commitment to user data privacy and compliance with relevant laws, with partnerships with vetted companies to prevent data misuse. Despite Reddit's concerns, Vana's DAO has a small user base compared to Reddit's vast community, with only 141,000 members.

How Neural Concept’s aerodynamic AI is shaping Formula 1

TechCrunch (April 14, 2024)

In 2018, Pierre Baqué caught the attention of researchers at the École Polytechnique Fédérale de Lausanne’s Computer Vision Laboratory, when after two days he came up with a design that closely approached what is now recognized as one of the most aerodynamic bike shapes. Six years later, Baqué is the CEO of Neural Concept, a startup that leverages an advanced version of the technology he pioneered to boost the performance of Formula 1 teams.

Neural Concept’s technology, Neural Concept Shape (NCS), is rooted in Computational Fluid Dynamics (CFD). It facilitates complex 3D aerodynamic simulations, allowing engineers to explore and refine aerodynamic models, making informed suggestions. This not only enhances race car performance by optimizing aerodynamics but also has applications in developing solutions like more energy-efficient battery cooling plates.

The company primarily targets the aerospace and automotive sectors, with a particular emphasis on Formula 1, where the demands are more pressing. Baqué, an avid F1 enthusiast from an early age, proudly discusses the significant performance improvements his technology has contributed to several teams, although confidentiality restrict the teams he can talk about. For instance, the Williams team, which is currently undergoing a rebuilding phase, employs NCS as a key tool in their quest to climb back to the top of the rankings.

The use of NCS significantly cuts down the time and cost associated with conventional CFD simulations. Formula 1 teams are charged between €100,000 to €1 million per year for using this technology, a cost-effective investment compared to their average annual spending of around $135 million.

The potential of AI in racing, from strategic planning to car setup, is increasingly recognized. Neural Concept is exploring new uses such as developing more efficient electric motors, optimizing vehicle cabin heating and cooling, and improving crash testing methods. A key focus is enhancing car crashworthiness while reducing excess weight. Currently, technological limits restrict simulations to individual components, setting the stage for future innovations. Following a $9.1 million Series A in 2022, the company is now progressing towards a Series B funding round.

Credits:

• Rafaella Andrade
• GANGA SINGH
• GDSC Events Team

References

1. Meta. (2024). Introducing Our Next-Generation Infrastructure for AI. Retrieved April 10, 2024, from https://about.fb.com/news/2024/04/introducing-our-next-generation-infrastructure-for-ai/
2. Reuters. (2024, April 10). Meta debuts new generation AI chip. Retrieved April 10, 2024, from https://www.reuters.com/technology/meta-debuts-new-generation-ai-chip-2024-04-10/
3. Lerman, R. (2024, April 9). AI data security startup Cyera confirms $300M raise at a $1.4B valuation. TechCrunch. Retrieved April 9, 2024, from https://techcrunch.com/2024/04/09/ai-data-security-startup-cyera-confirms-300m-raise-at-a-1-4b-valuation/
4. McFarland, M. (2024, April 13). Vana plans to let users rent out their Reddit data to train AI. TechCrunch. Retrieved April 13, 2024, from https://techcrunch.com/2024/04/13/vana-plans-to-let-users-rent-out-their-reddit-data-to-train-ai/
5. Bond, S. (2024, April 14). How Neural Concepts' aerodynamic AI is shaping Formula One. TechCrunch. Retrieved April 14, 2024, from https://techcrunch.com/2024/04/14/how-neural-concepts-aerodynamic-ai-is-shaping-formula-one/
6. O'Kane, S. (2024, April 7). Social order could collapse in AI era, two top Japan companies say. The Wall Street Journal. Retrieved April 8, 2024, from https://www.wsj.com/tech/ai/social-order-could-collapse-in-ai-era-two-top-japan-companies-say-1a71cc1d?st=61suw57mtqfb5cn
7. Alpert, L. (2024, April 11). A chip fab business grows in Brooklyn. The Wall Street Journal. Retrieved April 11, 2024, from https://www.wsj.com/articles/a-chip-fab-businessgrows-in-brooklyn-c5433a94?mod=ai_news_article_pos1
8. Sech, K. (2024, April 12). AI data centers' energy savings. The Wall Street Journal. Retrieved April 12, 2024, from https://www.wsj.com/business/energy-oil/ai-data-centers-energy-savings-d602296e