Computing and Hardware Archives - Page 3 of 7

Nvidia’s Explosive Q2 Performance Validates Generative AI Growth

Nvidia, the US-based semiconductor giant, made a significant impact yesterday with its outstanding performance in the second quarter of the fiscal year 2023-24. The company’s revenue for this quarter reached an impressive $13.51 billion, marking an astonishing 88 percent surge from the previous quarter and a remarkable 101 percent increase compared to the same period last year.

This remarkable growth exhibited by Nvidia serves as undeniable proof that the movement towards generative artificial intelligence and large language models is not merely a trend but a tangible reality. As a major player in the field, Nvidia stands as a powerhouse in producing the cutting-edge chip technology that drives generative AI.

Jensen Huang, the CEO of Nvidia who established the company in 1993, expressed, “We are witnessing the dawn of a new computing era.”

Following the announcement, Nvidia’s stock price soared by 6 percent, reflecting the market’s enthusiastic response. The company has also seen an extraordinary year-on-year surge of 422 percent in its net income, reinforcing its status as a trillion-dollar valuation enterprise. Looking forward, Nvidia anticipates a revenue of $16 billion for the upcoming fiscal year, with a potential margin of error of 2 percent.

Nvidia is poised for substantial growth in the forthcoming year, with the company’s optimistic outlook stemming from a clear increase in demand. To accommodate this rising demand, Nvidia has proactively secured an expanded supply.

The past year has undoubtedly been a momentous one for Nvidia. The launch of the GH200 Grace Hopper Superchip designed for intricate AI workloads, alongside the introduction of the Nvidia L40S GPU aimed at accelerating complex applications, underscored the company’s commitment to innovation. Notably, the upcoming GH200 Superchips, slated for release next year, will feature the same GPU as the highly sought-after H100 AI chip but with triple the memory capacity.

Huang further emphasized, “Throughout the quarter, major cloud service providers unveiled extensive NVIDIA H100 AI infrastructures. Prominent enterprise IT systems and software providers also formed partnerships to integrate NVIDIA AI across various industries. The race to adopt generative AI is in full swing.”

Nvidia’s chips have revolutionized graphics through AI, ushering in an entirely new realm of experience for gamers. The gaming sector has contributed significantly to Nvidia’s revenue, with a substantial $2.49 billion generated, marking an 11 percent increase from the first quarter and a notable 22 percent rise from the previous year.

Achieving a valuation exceeding a trillion dollars in May placed Nvidia among the elite group of US companies in this category. Investors have gravitated towards Nvidia, recognizing it as a major beneficiary of the AI surge. Even OpenAI, the catalyst behind the generative AI boom with its chatbot, utilized Nvidia’s H100 Hopper chips for training and running its GPT model.

Nvidia has emerged as a strong competitor to prominent GPU manufacturers based in China, Taiwan, and Hong Kong – key players in the industry. The US-imposed export regulations from October of the preceding year have placed Chinese firms at a disadvantage, as they are now restricted from purchasing US-manufactured chips. Simultaneously, the US government has invested substantial funds as incentives for the domestic chip sector.

Consequently, Nvidia’s strategic exports, such as the A800 processor to China, have inadvertently contributed to a dominant GPU market position. Huang affirmed, “Companies worldwide are making the shift from general-purpose computing to accelerated computing and generative AI.”

Nevertheless, China remains determined not to lag behind in the AI race. According to a Financial Times report, major Chinese entities like Baidu, ByteDance, Tencent, and Alibaba have placed significant orders with Nvidia, encompassing about 100,000 A800 processors and GPUs. The combined billing for these orders stands between $1 billion and $4 billion, showcasing China’s fervent pursuit of AI technology.

IBM Introduces Innovative Analog AI Chip That Works Like a Human Brain

IBM has taken the wraps off a groundbreaking analog AI chip prototype, designed to mimic the cognitive abilities of the human brain and excel at intricate computations across diverse deep neural network (DNN) tasks.

This novel chip’s potential extends beyond its capabilities. IBM asserts that this cutting-edge creation has the potential to revolutionize artificial intelligence, significantly enhancing its efficiency and diminishing the power drain it imposes on computers and smartphones.

Unveiling this technological marvel in a publication from IBM Research, the company states, “The fully integrated chip features 64 AIMC cores interconnected via an on-chip communication network. It also implements the digital activation functions and additional processing involved in individual convolutional layers and long short-term memory units.”

A Paradigm Shift in AI Computing

Fashioned within the confines of IBM Albany NanoTech Complex, this new analog AI chip comprises 64 analog in-memory compute cores. Drawing inspiration from the operational principles of neural networks within biological brains, IBM has ingeniously incorporated compact, time-based analog-to-digital converters into every tile or core. This design enables seamless transitions between the analog and digital domains.

Furthermore, each tile, or core, is equipped with lightweight digital processing units adept at executing uncomplicated nonlinear neuronal activation functions and scaling operations, as elaborated upon in an August 10 blog post by IBM.

A Potential Substitution for Existing Digital Chips

In the not-so-distant future, IBM’s prototype chip may very well take the place of the prevailing chips propelling resource-intensive AI applications in computers and mobile devices. Elucidating this perspective, the blog post continues, “A global digital processing unit is integrated into the middle of the chip that implements more complex operations that are critical for the execution of certain types of neural networks.”

As the market witnesses a surge in foundational models and generative AI tools, the efficacy and energy efficiency of conventional computing methods upon which these models rely are confronting their limits.

IBM has set its sights on bridging this gap. The company contends that many contemporary chips exhibit a segregation between their memory and processing components, consequently stymying computational speed. This dichotomy forces AI models to be stored within discrete memory locations, necessitating constant data shuffling between memory and processing units.

Drawing a parallel with traditional computers, Thanos Vasilopoulos, a researcher based at IBM’s Swiss research laboratory, underscores the potency of the human brain. He emphasizes that the human brain achieves remarkable performance while consuming minimal energy.

According to Vasilopoulos, the heightened energy efficiency of the IBM chip could usher in an era where “hefty and intricate workloads could be executed within energy-scarce or battery-constrained environments,” such as automobiles, mobile phones, and cameras.

He further envisions that cloud providers could leverage these chips to curtail energy expenditures and reduce their ecological footprint.

Amazon Advances in Generative AI with Custom Chips and Tools

While Microsoft and Google have held the spotlight in the field of generative AI, Amazon, quietly driven by its founder Jeff Bezos, has been making significant strides in enabling its customers to directly engage with this cutting-edge technology. Discreetly housed within an unassuming building in Austin, Texas, Amazon engineers are fervently crafting two distinct categories of microchips with the sole purpose of facilitating the training and execution of AI models, according to a report by CNBC.

The global stage saw the emergence of generative AI with the launch of OpenAI’s ChatGPT last year. Swift to react, Microsoft capitalized on its previous collaboration with OpenAI, seamlessly integrating the AI model’s capabilities into its existing products. Yet, this landscape is now poised for a substantial transformation in terms of use cases for this technology. Amazon, boasting a robust 40 percent share of the cloud computing market, attributes this transformative potential to the dearth of tools that empower businesses to harness their pre-existing data and effectively train models with it.

Executives from Amazon conveyed to CNBC that enterprises were reluctant to migrate their cloud data to Microsoft solely due to its leadership status in the realm of generative AI. Consequently, Amazon has chosen to invest its efforts in developing tools that empower businesses to leverage their cloud-stored data directly.

Amazon’s Tool Arsenal

Diverging from the trajectory of enabling users to deploy language models like GPT on its cloud servers, Amazon has forged its own collection of expansive language models, christened “Titan.” This suite of models comes complemented by an ancillary service called “Bedrock,” tailored for applications in generative AI. Bedrock not only provides users access to Amazon’s models but also grants entry to a spectrum of models devised by external entities such as Anthropic, Stability AI, and AI21 Labs.

However, Amazon’s innovation doesn’t end at providing its own Large Language Models (LLMs). The company remains cognizant of the fact that its LLM may not universally fit all use cases, and therefore aims to afford users the autonomy to select the model that optimally aligns with their specific application requirements.

Showcasing the Power of In-House Chips

Amazon appears to be mirroring the Silicon Valley trend, where companies are increasingly shunning traditional chip manufacturers in favor of crafting their own chips. Interestingly, this strategic shift isn’t new for Amazon. Almost a decade ago, the company integrated its custom-designed silicon, known as “Nitro,” into its cloud infrastructure. With over 20 million Nitro chips deployed—essentially one for each AWS server—Amazon has established a significant presence.

In 2018, Amazon unveiled the Graviton, an x86 chip competing with offerings from industry giants like AMD and Intel, designed specifically for Arm-based servers. Concurrently, Amazon embarked on the development of AI-focused chips, a move aimed at challenging Nvidia’s preeminence in this domain.

Branded as “Trainium” and “Inferentia,” Amazon’s chip offerings are named to underscore their roles in training and executing AI models, respectively. Inferentia, now in its second generation, prioritizes a cost-efficient, high-throughput framework for model execution. Meanwhile, Trainium delivers a remarkable 50 percent enhancement in price performance compared to alternative training model methods within the AWS ecosystem, as per insights shared by Amazon executives with CNBC. Confident in the appeal of its offerings, Amazon envisions companies opting for its chips to train their models rather than parting with data to OpenAI.

With an integrated infrastructure and a robust suite of tools at its disposal, Amazon is poised not only to catch up with the likes of Google and Microsoft but potentially to surpass them at an accelerated pace.

Nvidia Unveils GH200 GraceHopper: Next-Gen Superchips for Complex AI Workloads

In a recent press release, Nvidia, the world’s foremost supplier of chips for artificial intelligence (AI) applications, has introduced its latest breakthrough: the next generation of superchips, designed to tackle the most intricate generative AI workloads. This revolutionary platform, named GH200 GraceHopper, boasts an unprecedented feature: the world’s first HBM3e processor.

Combining Power: The Birth of GH200 GraceHopper

Nvidia’s ingenious GH200 GraceHopper superchip is the result of merging two distinct platforms: the Hopper platform, housing the graphic processing unit (GPU), and the Grace CPU platform, responsible for processing needs. These platforms, named in honor of computer programming pioneer Grace Hopper, have been seamlessly amalgamated into a singular superchip, paying homage to her legacy.

From Graphics to AI: The Evolution of GPUs

Historically, GPUs have been synonymous with high-end graphic processing in computers and gaming consoles. However, their immense computational capabilities have found new applications in fields like cryptocurrency mining and AI model training.

Powering AI through Collaborative Computing

Notably, Microsoft’s Azure and OpenAI have harnessed Nvidia’s chips to build substantial computing systems. By employing Nvidia’s A100 chips and creating infrastructures to distribute the load of large datasets, Microsoft facilitated the training of GPT models, exemplified by the popular ChatGPT.

Nvidia’s Pursuit of AI Dominance

Nvidia, the driving force behind chip production, now seeks to independently construct large-scale data processing systems. The introduction of the Nvidia MGX platform empowers businesses to internally train and deploy AI models, underscoring Nvidia’s commitment to AI advancement.

The GH200 GraceHopper: A Leap Forward in Superchip Technology

Nvidia’s achievement in crafting the GH200 superchip can be attributed to its proprietary NVLink technology, which facilitates chip-to-chip (C2C) interconnections. This innovation grants the GPU unfettered access to the CPU’s memory, resulting in a robust configuration that offers a substantial 1.2 TB of high-speed memory.

Unveiling the HBM3e Processor

The GH200 GraceHopper is distinguished by the inclusion of the world’s inaugural HBM3e processor, surpassing the computational speed of its predecessor, HBM3, by an impressive 50%. In a single server setup, featuring 144 Neoverse cores, a staggering eight petaflops of AI performance can be achieved. With a combined bandwidth of 10TB/sec, the GH200 platform possesses the capability to process AI models that are 3.5 times larger and 3 times faster than previous Nvidia platforms.

Nvidia’s Unrivaled Market Position

Having briefly entered the $1 trillion valuation echelon earlier in the year, Nvidia commands over 90% of the market share in chip supply for AI and related applications. The demand for GPUs extends beyond training AI models to their operational execution, and this demand is poised to escalate as AI integration becomes commonplace. Evidently, not only chip manufacturers such as AMD, but also tech giants like Google and Amazon, are actively developing their offerings in this burgeoning sector.

Charting a Technological Course: GH200’s Arrival

The unveiling of the GH200 GraceHopper superchip solidifies Nvidia’s status as the premier technology provider. Anticipated to be available for users in Q2 2024, these groundbreaking chips promise to reshape the landscape of AI processing, further establishing Nvidia’s dominance in the industry.

Transformative Research Program: Growing Human Brain Cells on Silicon Chips Unlocks Lifelong Learning for Advanced A

Monash University’s Associate Professor Adeel Razi, in collaboration with Melbourne start-up Cortical Labs, has received a prestigious grant of approximately AUD 600,000 from the National Intelligence and Security Discovery Research Grants Program, as per a press release.

This groundbreaking research initiative aims to push the boundaries of machine learning by growing human brain cells on silicon chips.

The core of the study revolves around cultivating around 800,000 living brain cells on silicon chips, with the goal of teaching these cells to perform specific tasks. The project builds upon a successful endeavor last year, where brain cells were taught to play the computer game Pong, garnering global attention and marking a significant milestone for the team.

Associate Professor Razi explained that this integration of lab-grown brain cells with silicon chips paves the way for “programmable biological computing platforms,” effectively merging artificial intelligence and synthetic biology. He believes that this technology could eventually outperform traditional silicon-based hardware.

The Future of Machine Intelligence

The key breakthrough of this research lies in its focus on “continual lifelong learning,” a capability lacking in current AI systems. The ability to learn throughout a machine’s lifetime, acquire new skills without forgetting previous ones, adapt to changes, and apply past knowledge to novel tasks while conserving computing power, memory, and energy represents a paradigm shift in the future of machine intelligence.

Unlike traditional AI, which suffers from “catastrophic forgetting,” where new data overwrites previously acquired information, the human brain excels at continual lifelong learning, allowing individuals to adapt and learn throughout their lives.

This new generation of machine learning applications, including self-driving cars, autonomous drones, delivery robots, and intelligent wearable devices, demands a different kind of intelligence—one that can continuously evolve and learn.

Unlocking the Secrets of Lifelong Learning

At the heart of the project lies the DishBrain system—an innovative laboratory dish where human brain cells are cultivated. The research team aims to gain a deeper understanding of the biological mechanisms underlying lifelong continual learning and replicate these mechanisms to create more advanced AI machines with remarkable capabilities.

Associate Professor Razi stated that the grant will be utilized to develop AI machines that can replicate the learning capacity of biological neural networks. The goal is to scale up the hardware and methods to the point where they become a viable replacement for conventional computing.

The impact of this research could provide Australia with a significant strategic advantage in various industries, propelling the country to the forefront of the AI revolution. The project is highly anticipated, with its potential to unlock the secrets of lifelong learning as it progresses.

Unveiling the Condor Galaxy-1: The World’s Fastest Supercomputer Revolutionizes AI Model Training

Abu Dhabi’s technology holding group, G42, has recently unveiled the Condor Galaxy-1 (CG-1), the world’s fastest supercomputer, boasting an incredible 54 million cores and a processing capacity of four exaflops. The CG-1 is strategically located in Santa Clara, California, and will be operated by the US-based AI firm, Cerebras, ensuring compliance with US laws.

With the increasing prominence of artificial intelligence (AI) technology, there is a growing demand for supercomputers to facilitate the training of sophisticated AI models. Companies like Microsoft have offered their services to construct and rent out such costly infrastructure to other businesses seeking AI solutions.

G42, headquartered in Abu Dhabi, is a technology holding group that envisions creating infrastructure for a futuristic world. Collaborating with both nations and corporations, G42 aims to solve humanity’s most pressing problems through innovative initiatives, including the development of the Condor Galaxy system of supercomputers.

What Sets the Condor Galaxy System Apart? While many technology companies have opted for massive clusters of graphic processing units (GPUs) for AI models, Cerebras Systems’ CEO, Andrew Feldman, and his team at Cerebras, working with G42, have taken a different approach. They are building an interconnected set of AI supercomputers that significantly reduce the time required for AI model training. In fact, the process of setting up generative AI models using their approach can be accomplished within minutes by a single individual, as opposed to the months and rare expertise typically needed when relying on thousands of tiny GPUs.

The combined processing power of the Condor Galaxy System will reach an astonishing 36 exaFLOPS, setting a new benchmark in the realm of computing.

The Core of the Condor Galaxy System – CG-1:

At the heart of the Condor Galaxy System lies the CG-1 supercomputer, which has recently been unveiled. Cerebras has meticulously assembled 64 of its flagship CS-2 AI processors to create the CG-1 supercomputer. AMD’s EPYC processor cores power the system, boasting an impressive 54 million AI-optimized compute cores, 388 terabits per second of fabric bandwidth, and 82 TB of memory storage.

Operating at 16-bit computation, CG-1 achieves a remarkable four exaFLOPS of computing power, outpacing the fastest supercomputer currently in existence by a factor of four. Its capabilities extend to training 600 billion parameter models and can be further expanded to support 100 trillion parameter models, far surpassing the parameters required for OpenAI’s GPT-4 model, which employs 1.7 trillion parameters.

Future Plans and Accessibility:

Cerebras and G42 have ambitious plans to introduce two additional installations of the supercomputer, CG-2 and CG-3, in the US in early 2024, while also offering CG-1 as a cloud service to customers.

What sets CG-1 apart is its native compatibility with 50,000 tokens and the ability to function without the need for complex distributed programming languages or special software libraries. This unique feature saves valuable time that would otherwise be spent on distributing workflows across multiple GPUs.

Location and Security Measures:

The supercomputer is housed at Colovore, a colocation facility in Santa Clara, California. It will be operated by Cerebras under the jurisdiction of US laws, ensuring that its computing power remains inaccessible to adversarial nations.

A Vision for the Future:

G42 and Cerebras are confident that the unmatched capabilities of the world’s fastest supercomputer will enable significant progress in addressing critical challenges in healthcare, energy, and climate change. The Condor Galaxy-1 stands as a testament to the potential of supercomputing technology in shaping a better future for humanity.

Microsoft Unveils the World’s First Analog Optical Computer to Solve Optimization Problems

Microsoft Research Lab in Cambridge has unveiled the world’s first analog optical computer which promises to solve optimization problems at a lightning-fast pace, a press release said. The computer uses photons and electrons to process continuous value data instead of crunching them to binary bits using transistors.

Optimization problems are everywhere around us whether one considers managing electricity on the grid or delivering goods to your doorstep from the warehouse of the seller. Optimizing involves the use of the least resources to maximize returns for processes. However, even the world’s fastest computers can end up spending years to solve them once the size of the problem grows.

The Traveling Salesman Problem is a classic example of this problem. It involves finding an optimum route to visit a set of cities just once before returning to the starting point. When computing for five cities, there are 12 possible routes that one can take. However, as the numbers of cities grow, the potential routes expand exponentially making them impossible to compute.

The Analog Iterative Machine

Researchers have used heuristic algorithms which can provide approximate solutions to such problems. However, even with their custom hardware, the approach has not yielded a practical alternative to conventional computers, which are limited by their binary abstraction of problems.

The research team at Microsoft suggests a more expressive abstraction that allows the use of mixed variables, both binary and continuous to solve problems of optimization. The team achieved this using an analog optical computer that they call the Analog Interactive Machine (AIM).

The team leverages the ability of photons to not interact with each other but with the matter through which they travel to perform simple mathematical operations like addition and multiplication.

By constructing a physical system that uses optics and electronics to perform vector-matrix multiplications, the team has found a way to efficiently and swiftly execute calculations needed to find solutions to optimization problems.

Further, the components of this system have been miniaturized to fit tiny centimeter-scale chips, making the AIM no bigger than a rack enclosure.

Real-world applications

Last year, the company built the first generation AIM computer that delivered an accuracy of up to seven bits. Now in an attempt to test it in the real world, Microsoft has teamed up with Barclays, a UK-based bank to test applications in financial markets.

Interbank transactions are settled at clearing houses which process hundreds of thousands of transactions on a daily basis. As banking transactions scale, the settlements take increasingly longer to be completed which is a real-world optimization problem.

The Microsoft team has already attempted using a basic version of AIM to solve the transaction problem and solved with accurately in tests so far. The team is now working to scale up the computer to handle a larger number of variables and more data.

Microsoft believes that an optical computer can address the two major issues with silicon-based computing. First, the diminishing returns on Moore’s Law where computing capacity per dollar has been declining over the years in chips as well as the limitations of computing in binary.

An optical computer could literally open a spectrum of options for researchers while also reducing resources spent on performing complex calculations.

Microsoft First Step To A Quantum Supercomputer

Microsoft Has Completed The First Step To Building A Quantum Supercomputer

Microsoft is leading the race in artificial intelligence (AI) models and has also set its eye on the future of computing. In an announcement made on Wednesday, the Redmond, Washington-headquartered company unveiled a roadmap where it plans to build a quantum supercomputer in the next 10 years.

Quantum computing has been in the news in recent weeks for beating supercomputers at complex math and being able to compute at speeds much faster than one could imagine. Scientists have acknowledged that they have used noisy physical qubits for these achievements, which are not error-free.

Microsoft refers to today’s quantum computers as those belonging to the foundational level. According to the software giant, these computers need upgrades in the underlying technology, much like early computing machines did as they moved from vacuum tubes to transistors and then to integrated circuits before taking their current form.

Logical qubits

In its roadmap, Microsoft suggests that as an industry, quantum computing needs to move on from noisy physical qubits to reliable logical qubits since the former cannot reliably run scaled applications.

Microsoft suggests bundling hundreds to thousands of physical qubits into one logical qubit to increase redundancy and reduce error rates. Since qubits are prone to interference from their environment, efforts must be made to increase their stability, which will aid in increasing their reliability.

Reliable logical qubits can be scaled to perform complex problems that need solving urgently. However, since we do not have a measure of how reliable calculations in quantum computing are, the company has proposed a new measure called reliable Quantum Operations Per Second (rQOPS) to do so.

Microsoft claims that the Majorana-based qubit announced last year is highly stable but also difficult to create. The company has published its progress in the peer-reviewed publication in the journal Physical Review B.

Platform to accelerate discovery

Microsoft has completed the first step to building a quantum supercomputer — When quantum computing will reach the supercomputer stageMicrosoft

Microsoft estimates that the first quantum supercomputer will need to deliver at least one million rQOPS with an error rate of 10^-12,or one in every trillion operations, to be able to provide valuable inputs in solving scientific problems. However, quantum computers of today deliver an rQOPS value of zero, meaning that the industry as a whole has a long way to go before we see the first quantum supercomputer.

Instead of decades, Microsoft wants to build this supercomputer in a matter of years and has now launched its Azure Quantum Elements platform to accelerate scientific discovery. The platform will enable organizations to leverage the latest breakthroughs in high-performance computing (HPC), AI, and quantum computing to make advances in chemistry and material science to build the next generation of quantum computers.

The company is also extending its Copilot services to Azure Quantum, where researchers will be able to use natural language processing to solve complex problems of chemistry and materials science. Copilot can help researchers query quantum computers and visualize data using an integrated browser.

Microsoft’s competitors in this space are Google and IBM, who have also unveiled their quantum capabilities.

Opaque Systems Confidential Computing Platform

Opaque Systems Introduces Innovations for Confidential Computing Platform

Opaque Systems, an AI and analytics company, has unveiled new advancements for its confidential computing platform, prioritizing the confidentiality of organizational data in conjunction with large language models (LLMs).

The company’s latest offerings will be showcased during Opaque’s keynote address at the inaugural Confidential Computing Summit, scheduled for June 29 in San Francisco.

Among the innovations are a privacy-preserving generative AI optimized for Microsoft Azure’s Confidential Computing Cloud and a zero-trust analytics platform called Data Clean Room (DCR). Opaque Systems integrates secure hardware enclaves and unique cryptographic fortifications to provide multiple layers of protection for its generative AI, ensuring data remains encrypted throughout model training, fine-tuning, and inference stages.

Jay Harel, VP of product at Opaque Systems, emphasized the platform’s commitment to data security, stating, “Our platform safeguards data at rest, in transit, and while in use, minimizing the likelihood of data breaches throughout the lifecycle.”

These advancements aim to enable organizations to securely analyze confidential data while ensuring its confidentiality and safeguarding against unauthorized access. Opaque achieves this by executing machine learning and AI models on encrypted data within trusted execution environments (TEEs), preventing unauthorized access to sensitive information.

Furthermore, Opaque’s Data Clean Rooms (DCRs) operate on the principle of zero-trust, encrypting data at rest, in transit, and during usage. This comprehensive approach ensures that data remains confidential throughout the entire process.

To fully unleash the potential of LLMs like ChatGPT, Opaque highlights the need to train these models on confidential data without the risk of exposure. The company recommends adopting confidential computing, which safeguards data throughout the model training and inference process, unlocking the transformative capabilities of LLMs.

Opaque utilizes Confidential Computing technology to leverage specialized hardware provided by cloud providers. By encrypting datasets end-to-end throughout the machine learning lifecycle, Opaque’s platform ensures the privacy of the model, prompt, and context during training and inference.

Harel identified three main issues concerning generative AI and privacy, particularly with LLMs:

Queries: LLM providers have visibility into user queries, potentially accessing sensitive information such as proprietary code or personally identifiable information (PII). This concern is amplified by the increasing risk of hacking.
Training models: Providers often access and analyze internal training data to enhance AI models. However, retaining training data can accumulate confidential information, heightening vulnerability to data breaches.
IP issues for organizations with proprietary models: Fine-tuning models using proprietary data requires granting LLM providers access to sensitive information or deploying proprietary models within the organization. This exposes private and sensitive data to external individuals, raising the risk of hacking and data breaches.

Opaque Systems has developed its generative AI technology with these concerns in mind, aiming to facilitate secure collaboration among organizations and data owners while ensuring regulatory compliance. With Opaque’s platform, organizations can train, fine-tune, and run inference on LLMs without gaining direct access to the raw data itself, preserving data privacy.

The company’s Data Clean Room (DCR) offering is reinforced with secure hardware enclaves and cryptographic fortification, delivering multiple layers of protection against cyberattacks and data breaches. Operating within a cloud-native environment, the system executes within a secure enclave on the user’s cloud instance, allowing businesses to retain their existing data infrastructure.

Harel emphasized Opaque’s mission to prioritize the privacy of confidential data, stating, “For AI workloads, we enable businesses to keep their data encrypted and secure throughout the lifecycle, significantly reducing the likelihood of loss. Data is kept confidential at rest, in transit, and while in use.”

Opaque Systems innovative advancements in confidential computing provide organizations with robust solutions for data privacy and secure analytics, empowering them to leverage AI technologies while maintaining the confidentiality of their data.

Chinese Chips Have Made Their Way Into US Government Agencies

It seems that dubious Chinese encryption chips have made their way into the U.S. government agencies and international military organizations, as per a report by Wired.

These encryption keys, with their advanced algorithms, protect the confidential information in possession of organizations like the North Atlantic Treaty Organization (NATO), NASA, the U.S. Navy, and the U.K. military – all reportedly using encryption microcontroller chips supplied by Chinese chipmaker Hualan Microelectronics, also known as Sage Microelectronics.

On the Entity List

Hualan was added to the U.S. ‘Entity List’ in 2021 by the Commerce Department’s Bureau of Industry and Security (BIS). The Entity List enlists all the companies the U.S. has placed sanctions on. Hualan had been added to the list for “acquiring and … attempting to acquire US-origin items in support of military modernization for [China’s] People’s Liberation Army.”

The U.S. has been at odds with Chinese companies that have close relations with the Chinese Communist Party (CPP), which includes virtually all companies on the mainland. Exhibit A: TikTok. Other companies that have been put on the sanction list include Huawei, AI startup SenseTime, drone supplier DJI, etc., as the U.S. becomes increasingly stringent about its policies in regard to Chinese technology.

Federal procurement records show that U.S. government agencies, from the Federal Aviation Administration to the Drug Enforcement Administration and the U.S. Navy, have bought encrypted hard drives that use the chips, too, said the Wired report.

A disconnect between U.S. government wings

The disconnect between the Bureau and government agencies may be because the chips were supplied by Initio, a subsidiary of Hualan. Initio was acquired by Hualan in 2016 and is headquartered in Taiwan. The chips have the Initio branding.

The Wired report further says that the Chinese may have a hidden backdoor that would allow China’s government to stealthily decrypt Western agencies’ secrets. And while no such backdoor has been found, security researchers warn that if one did exist, it would be virtually impossible to detect.

“If a company is on the Entity List with a specific warning like this one, it’s because the U.S. government says this company is actively supporting another country’s military development,” said Dakota Cary, a China-focused research fellow at the Atlantic Council, a Washington, DC-based think tank, in an interview with Wired. “It’s saying you should not be purchasing from them, not just because the money you’re spending is going to a company that will use those proceeds in the furtherance of another country’s military objectives, but because you can’t trust the product.”

“It’s used somewhat as a blacklist,” said Emily Weinstein, a researcher at Georgetown University’s Center for Security and Emerging Technology, while speaking to Wired. “The Entity List should be a red or maybe a yellow alert to anyone in the US government who’s working with this company to take a second look at this.”

A spokesperson with the Bureau said that although a company like Initio – an unlisted subsidiary – isn’t technically affected by the Entity List, “as a general matter, affiliation with an Entity Listed party should be considered a ‘red flag.’”

With suspicious Chinese balloons flying over U.S. airspace and new allegations of Chinese apps spying on U.S. citizens, the diplomatic relations between the two countries have taken a hit. The latest Wired report adds to the flare-up.