Cloud Hosting

OAK RIDGE, Tenn. At Oak Ridge National Laboratory, the government-funded science research facility nestled between Tennessees Great Smoky Mountains and Cumberland Plateau that is perhaps best known for its role in the Manhattan Project, two supercomputers are currently rattling away, speedily making calculations meant to help tackle some of the biggest problems facing humanity.

You wouldnt be able to tell from looking at them. A supercomputer called Summit mostly comprises hundreds of black cabinets filled with cords, flashing lights and powerful graphics processing units, or GPUs. The sound of tens of thousands of spinning disks on the computers file systems, and air cooling technology for ancillary equipment, make the device sound somewhat like a wind turbine and, at least to the naked eye, the contraption doesnt look much different from any other corporate data center. Its next-door neighbor, Frontier, is set up in a similar manner across the hall, though its a little quieter and the cabinets have a different design.

Yet inside those arrays of cabinets are powerful specialty chips and components capable of, collectively, training some of the largest AI models known. Frontier is currently the worlds fastest supercomputer, and Summit is the worlds seventh-fastest supercomputer, according to rankings published earlier this month. Now, as the Biden administration boosts its focus on artificial intelligence and touts a new executive order for the technology, theres growing interest in using these supercomputers to their full AI potential.

The more computation you use, the better you do, said Neil Thompson, a professor and the director of the FutureTech project at MITs Initiative on the Digital Economy. Theres this incredibly predictive relationship between the amount of computing you use in an AI system and how well you can do.

At the department level, the new executive order charges the Department of Energy with creating an office to coordinate AI development among the agency and its 17 national laboratories, including Oak Ridge. Critically, the order also calls on the DOE to use its computing and AI resources for foundation models that could support climate risk preparedness, national security and grid resilience, among other applications which means increased focus on systems like Frontier and Summit.

The executive order provided us clear direction to, first of all, leverage our capabilities to make sure that we are making advances in AI, but were doing it in a trustworthy and secure way, said Ceren Susut, the DOEs associate director of science for Advanced Scientific Computing Research. That includes our expertise accumulated in the DOE national labs, and the workforce, of course, but also the compute capabilities that we have.

The governments AI specs

Supercomputers like Summit and Frontier can be measured in performance. Often, theyre measured in exaflops, defined as their ability to calculate a billion a billion no, this isnt a typo floating point operations per second. Frontier sits at 1.194 exaflops, while Summits is a little less impressive, at 148.60 petaflops. But they can also be measured in their number of GPUs: Summit has slightly more than 28,000, while Frontier has nearly 10,000 more.

These chips are particularly helpful, experts explain, for the types of matrix algebra calculations one might need for training AI models. Notably, the DOE is nearing the completion of its Exascale Computer Project, an initiative across the national labs to rewrite software to be GPU, or AI, enabled. Many of these applications are integrating AI techniques as one way in which they take advantage of GPUs, Susut told FedScoop in an email.

In the same vein, one of the biggest requirements for building advanced AI systems, including AI tools developed with the help of the government, has become compute, or computational resources. For the same reason, the technical needs of the most powerful supercomputers, and most rigorous AI models, can often line up. Thats where systems like Frontier and Summit come in.

Ive read so many papers recently about how AI and [machine learning] need high bandwidth, low latency, high-performance networks around high memory nodes that have really fast processors on them, said Bronson Messer, the director of science at the Oak Ridge Leadership Computing Facility, which houses the two supercomputers. Im like, wow, thats exactly what Ive always wanted for 20 years.

MITs Thompson noted that in the field of computer vision, about 70 percent of the improvements in these systems can be attributed to increased computing power.

There are already efforts to train AI models, including large language models, at the lab. So far, researchers at Oak Ridge have used the labs computing resources to develop a machine learning algorithm designed to create simulations to boost greener flight technology; an algorithm to study potential links between different medical problems based on scans of millions of scientific publications; and datasets reflecting how molecules might be impacted by light information that could eventually be used for medical imaging and solar cells applications.

Theres also a collaboration with the National Cancer Institute focused on building a better way of tracking cancer across the country, based on a large dataset, sometimes called a corpus, of medical documents.

We end up with something on the order of 20 to 30 billion tokens, or words, within the corpus, said John Gounley, a computational scientist at Oak Ridge working on that project. Thats something where you can start legitimately training a large language model on a dataset thats that large. So thats where the supercomputer really comes in.

More AI initiatives at the facility will soon go online. The DOEs support for the Summit supercomputer has been extended, in part, to propel the National Artificial Intelligence Research Resource, which aims to improve government support for AI research infrastructure. Starting next year, several projects focused on building foundation models are set to start on Frontier, including an initiative that plans to use a foundation model focused on energy storage and a large language model built with a Veterans Affairs data warehouse.

How DOE pivots for the AI era

As part of the executive order, the Department of Energy is charged with building tools to mitigate AI risks, training new AI researchers, investigating biological and environmental hazards that could be caused by AI, and developing AI safety and security guidelines. But the agency doesnt need to be pushed to invest in the technology.

This past summer, the DOE disclosed around 180 public AI use cases as part of a required inventory. Energy is also working on preliminary generative AI programs, including new IT guidance and a specified Discovery Zone, a sandbox for trying out the technology. Earlier this year, the Senate hosted a hearing focused specifically on the DOEs work with the technology, and the agencys office of science has requested more resources to support its work on AI, too.

But as the agency looks to deploy supercomputers for AI, there are challenges to consider. For one, the increased attention toward the technology marks a significant pivot for the supercomputing field, according to Paola Buitrago, the director of artificial intelligence and data at the Pittsburgh Supercomputing Center. Traditionally, research on supercomputers has focused on topics like genomics and computational astrophysics research that has different requirements than artificial intelligence, she explained. Those limitations arent just technical, but apply to talent and workforce as well.

Most of the power of the impressive supercomputers could not always be leveraged completely or efficiently to service the AI computing needs, Buitrago said in an email. There is a mindset in the supercomputing field that doesnt completely align with what is needed to advance AI.

And the government only has so many resources. While there are several supercomputers distributed across some of the national labs, Oak Ridge itself can only support so much research at a time. Lawrence Berkeley National Laboratorys supercomputer might handle several hundred projects in a year, but Messer said Frontier and Summit have a smaller number of projects than other labs because the projects tend to run significantly longer.

Theres also more demand for supercomputing facilities than supply. Only a fraction of projects proposed to Oak Ridge are accepted. Meanwhile, while training foundation modelsis incredibly computationally demanding and only the largest supercomputers support developing them, building these systems is just one of several priorities that the agency must consider.

DOE is actively considering these ideas and must also balance the use of our supercomputers across a range of high-priority mission applications, said Susut, the DOE supercomputer expert. Our supercomputers are open to the research community through merit-based competitive allocation programs, and we have a wide diversity of users.

Even as the Department of Energy plans potential successors to Frontier, MITs Thompson noted that there are still other hurdles ahead.

For one, theres a tradeoff between the flexibility of these computers and efficiency, especially as the agency seeks even greater performance. Supercomputers, of course, are extremely expensive systems and costs arent dropping as fast as they used to. And they take time to build. At Oak Ridge, plans for a new computer, which will have AI as a key area of focus, are already in the works. But the device isnt expected to go online until 2027.

The reality is that the U.S. private sector has led research in AI starting in the past few years, as they have the data, the computing capacity and the talent, Buitrago said. Whether or not that continues to be the case depends on how much the government prioritizes AI and its needs. To extend, some may say the government is slowly catching up.

Follow this link:

How one national lab is getting its supercomputers ready for the AI age - FedScoop

Engineering | News releases | Technology

November 29, 2023

University of Washington researchers found that when prompted to create pictures of a person, the AI image generator over-represented light-skinned men, sexualized images of certain women of color and failed to equitably represent Indigenous peoples. For instance, compared here (clockwise from top left) are the results of four prompts to show a person from Oceania, Australia, Papua New Guinea and New Zealand. Papua New Guinea, where the population remains mostly Indigenous, is the second most populous country in Oceania.University of Washington/Stable Diffusion AI GENERATED IMAGE

What does a person look like? If you use the popular artificial intelligence image generator Stable Diffusion to conjure answers, too frequently youll see images of light-skinned men.

Stable Diffusions perpetuation of this harmful stereotype is among the findings of a new University of Washington study. Researchers also found that, when prompted to create images of a person from Oceania, for instance, Stable Diffusion failed to equitably represent Indigenous peoples. Finally, the generator tended to sexualize images of women from certain Latin American countries (Colombia, Venezuela, Peru) as well as those from Mexico, India and Egypt.

The researchers will present their findings Dec. 6-10 at the 2023 Conference on Empirical Methods in Natural Language Processing in Singapore.

Its important to recognize that systems like Stable Diffusion produce results that can cause harm, said Sourojit Ghosh, a UW doctoral student in the human centered design and engineering department. There is a near-complete erasure of nonbinary and Indigenous identities. For instance, an Indigenous person looking at Stable Diffusions representation of people from Australia is not going to see their identity represented that can be harmful and perpetuate stereotypes of the settler-colonial white people being more Australian than Indigenous, darker-skinned people, whose land it originally was and continues to remain.

To study how Stable Diffusion portrays people, researchers asked the text-to-image generator to create 50 images of a front-facing photo of a person. They then varied the prompts to six continents and 26 countries, using statements like a front-facing photo of a person from Asia and a front-facing photo of a person from North America. They did the same with gender. For example, they compared person to man and person from India to person of nonbinary gender from India.

The team took the generated images and analyzed them computationally, assigning each a score: A number closer to 0 suggests less similarity while a number closer to 1 suggests more. The researchers then confirmed the computational results manually. They found that images of a person corresponded most with men (0.64) and people from Europe (0.71) and North America (0.68), while corresponding least with nonbinary people (0.41) and people from Africa (0.41) and Asia (0.43).

Likewise, images of a person from Oceania corresponded most closely with people from majority-white countries Australia (0.77) and New Zealand (0.74), and least with people from Papua New Guinea (0.31), the second most populous country in the region where the population remains predominantly Indigenous.

A third finding announced itself as researchers were working on the study: Stable Diffusion was sexualizing certain women of color, especially Latin American women. So the team compared images using a NSFW (Not Safe for Work) Detector, a machine-learning model that can identify sexualized images, labeling them on a scale from sexy to neutral. (The detector has a history of being less sensitive to NSFW images than humans.) A woman from Venezuela had a sexy score of 0.77 while a woman from Japan ranked 0.13 and a woman from the United Kingdom 0.16.

We werent looking for this, but it sort of hit us in the face, Ghosh said. Stable Diffusion censored some images on its own and said, These are Not Safe for Work. But even some that it did show us were Not Safe for Work, compared to images of women in other countries in Asia or the U.S. and Canada.

While the teams work points to clear representational problems, the ways to fix them are less clear.

We need to better understand the impact of social practices in creating and perpetuating such results, Ghosh said. To say that better data can solve these issues misses a lot of nuance. A lot of why Stable Diffusion continually associates person with man comes from the societal interchangeability of those terms over generations.

The team chose to study Stable Diffusion, in part, because its open source and makes its training data available (unlike prominent competitor Dall-E, from ChatGPT-maker OpenAI). Yet both the reams of training data fed to the models and the people training the models themselves introduce complex networks of biases that are difficult to disentangle at scale.

We have a significant theoretical and practical problem here, said Aylin Caliskan, a UW assistant professor in the Information School. Machine learning models are data hungry. When it comes to underrepresented and historically disadvantaged groups, we do not have as much data, so the algorithms cannot learn accurate representations. Moreover, whatever data we tend to have about these groups is stereotypical. So we end up with these systems that not only reflect but amplify the problems in society.

To that end, the researchers decided to include in the published paper only blurred copies of images that sexualized women of color.

When these images are disseminated on the internet, without blurring or marking that they are synthetic images, they end up in the training data sets of future AI models, Caliskan said. It contributes to this entire problematic cycle. AI presents many opportunities, but it is moving so fast that we are not able to fix the problems in time and they keep growing rapidly and exponentially.

This research was funded by a National Institute of Standards and Technology award.

For more information, contact Ghosh at ghosh100@uw.edu and Caliskan at aylin@uw.edu.

Here is the original post:

AI image generator Stable Diffusion perpetuates racial and ... - University of Washington

Excitement and cautious optimism about the possibilities, but we must ensure that we harness AI equitably, ethically, and safely.

By Daniela Weber, Deputy Director of the NetHope Center for the Digital Nonprofit

AI has been a topic of NetHope Global Summit sessions since 2018, when a few of us (including myself, a Member CIO at the time) got into a room to discuss about the ethics of AI, and decided to start an AI workstream in NetHopes Emerging Technologies working group, as a vehicle to discuss how we might use AI in our sector, and how we could address those challenges. A number of NetHope Members have, since then, started using AI - piloting, and prototyping first; and some of these solutions are now scaling up and achieving real benefits.

At this years NetHope Global Summit, 12 dedicated AI sessions, with even more sessions that incorporated AI into the discussions, were included in the agenda and were standing room only. Clearly, with the arrival of generative AI in the mainstream, the discourse about the usefulness of AI, as well as its risks, has reached a new level of interest and has also been elevated to the leadership teams and boards of nonprofit organizations. The NetHope Membership AI workstream is continuing the conversations on the various aspects of AI and working together to identify opportunities for collective action. As an example, work on the Humanitarian AI Code of Conduct is continuing which was begun at the NetHope Global Summit.

I saw a very similar picture at this years Technology Association of Grantmakers (TAG) conference in Nashville: Keynotes, presentations, and many discussions on AI. I am very excited about the AI Adoption Framework for Philanthropy which we test-drove as a draft version during the conference. Version 1.0 will be launched on December 6th.

My takeaways from all of it:

There is immense interest in AI and what it can bring to the sector. From anticipatory action triggered by extreme weather forecasts or people movement predictions, to detecting sight-threatening conditions and providing care recommendations, to supporting conservation and crisis response efforts there are numerous ways in which AI can enable nonprofit organizations to deliver their mission and increase their impact in a complex, poly-crisis world. And particularly with generative AI we are starting to see how AI can bring more efficiency to areas such as fundraising, IT, Finance, and other secondary processes. This is badly needed at a time when nonprofit organizations are forced to do more with less given the resource constraints and lack of critical funding. We do need more solutions that are truly relevant and useful for the sector, and the capabilities to implement and use them at scale.

There is also immense concern about the risk areas of AI such as bias, data privacy, fake information, fraud, cybersecurity, the equitable access to the use of AI solutions as well as the longer-term societal impact. Many of our discussions during Summit were about how we can manage these risks and ensure we use AI responsibly. At minimum, nonprofit organizations need to have the governance in place to put guardrails around the use of AI tools, and solid cross-functional mechanisms to risk assess and decide when to move forward, and when to pause or stop. Toolkits such as NetHopes Artificial Intelligence (AI) Ethics for Nonprofits, and the soon to be released Gender Equitable AI toolkit, will help nonprofits to ensure their implementation and use of AI is ethical and equitable. Ideally though, the ecosystem of nonprofits, implementation partners, and funders will align behind a common framework on AI usage.

There are very practical roadblocks for the use of AI a lot of them linked to a lack of capacity and capability in many nonprofits, and the multitude of solutions and approaches to choose from, with roadmaps and pricing mechanisms that are not clearly defined yet, making the selection of the right solution a challenge. We need to engage with our technology partners to get that clarity so organizations can plan their next moves. The recent events and staff transitions at companies in the space have raised concerns not only about the differing viewpoints with regards to AI safety, but also the viability and stability of a key vendors in the generative AI space that many organizations already work with or are exploring. Vendor risk management remains a critical component of IT and procurement governance, as it should be for all technology solutions.

The old truth remains you need to have good data as the basis of your AI solution if you want it to produce useful output. A good data governance structure is essential to achieve this. NetHopes Data Governance Toolkit was created to support nonprofits with setting up this structure. There is also a lively discussion in our AI Working Group about the potential of sharing datasets between organizations and maybe we can go one step further and use all that great data we are collecting between us to train a shared large language model.

I, for one, am excited and cautiously optimistic about the possibilities but we must ensure that we harness AI equitably, ethically, and safely.

NetHopes New Strategic Plan Digital Transformation & Innovation Generative AI Program Highlight

NetHope has launched its 2024-2030 strategy. A generation ago, the Digital Divide described the gap in access first to computers, and then to the internet and essential technologies. But today, new digital challenges threaten international NGO mission advancement as organizations are forced to grapple with the widening gap between available resources and growing needs. These new challenges what NetHope calls the new Digital Divides represent critical gaps in infrastructure, systems, capabilities, and knowledge. Over the next decade, will lead collective action to bridge the New Digital Divides skills and leadership, inclusion, protection, transformation and innovation, and climate adaptation and resilience - leveraging our powerful, proven model for advance global good.

Bridging the New Digital Divides Spotlight: Digital Transformation and InnovationThe innovation required to meet emerging and existing challenges to the nonprofit sector and the communities we serve only happens when it is made a priority.NetHope works within our network, data, and history of catalyzing collective action to see complex problems are met with insights, policies, and advice that are context-appropriate, efficient and scalable. NetHope will use our influence, networks, data, and history of catalyzing new and high-potential ideas to ensure complex problems are met with insights, policies, guidance, and technology solutions.

Program Spotlight: Augmenting Generative AI Skills and Practices for NonprofitsNetHopes work in AI dates back to 2018 as much of our work begins, a gathering of Members and collective desire for action. Now, our program focuses on generative AI skills and practice building addresses the unique needs of nonprofits, their staff, and implementation partners. Training, as well as market scanning for sector appropriate solutions, and the exchange of knowledge and experiences are important elements of this program. NetHope has also published several resources on AI. Check out NetHopes recent Strategic Conversation: AI Ethics Today along with resources such as NetHopes Guide to Usefulness of Existing AI Solutions in Nonprofit Organizations, The AI Suitability Toolkit for Nonprofits, and the Artificial Intelligence Ethics for Nonprofits Toolkit.

Read the original here:

Amplifying the efforts of nonprofit organizations with AI - World - ReliefWeb

NEW YORK, Nov 29 (Reuters Breakingviews) - Artificial intelligence is spawning some genuinely concerning financial decisions. As technology titans jockey to back hot new startups, they are extracting explicit or implicit promises of revenue in return. The chaotic week of upheaval at OpenAI, the Microsoft-backed (MSFT.O) owner of ChatGPT, provides a window into how these sorts of supposedly virtuous circles have a way of turning vicious.

Mania over programs that emulate, or even surpass, human abilities is in full throttle. After OpenAI within days fired and rehired boss Sam Altman, following some intervention from Microsoft Chief Executive Satya Nadella, the startup looks to be pressing ahead with a planned share sale that could value it at more than $80 billion, according to the Financial Times. A deal would signal that the hype is strong enough to overcome the significant risks just laid bare, partly because of the corporate buttresses. Before the boardroom coup, OpenAI paused subscriptions for its paid ChatGPT Plus service because it couldnt keep up with the surge of new users. This level of exuberance evokes the dot-com era in more ways than one.

Microsofts injection of $10 billion into OpenAI in January helped kick off the craze. As part of their deal, the ChatGPT operator agreed to exclusively use its new investors cloud computing services. In the 12 months before they struck the arrangement, OpenAI had spent less than $1 million with the software goliath, according to tech news outfit The Information, with the figure now on pace to exceed $400 million a year on Microsofts Azure alone.

So-called generative AI, which uses algorithms to produce new text or imagery from the data on which it is trained, has released animal spirits across Silicon Valley. Microsoft, Apple (AAPL.O) and Alphabet (GOOGL.O) are among those handily outperforming the Nasdaq Composite Index (.IXIC) this year. The excitement makes some sense, but there is notable danger embedded in the way the investments are being structured, just as there was in the 1990s when stretching communications networks across the globe was all the rage. Both turbocharge new ideas in ways that simultaneously manufacture fresh demand for existing products. But regulation or other complications could easily slow progress or create extra hurdles for AI adoption, in turn jeopardizing all the extra revenue Big Tech is anticipating and their ballooning valuations.

OpenAIs recent fundraising could make it worth more than double the $30 billion imputed earlier this year. The uplift has been symbiotic. Microsoft, including debt, is now valued at about $2.8 trillion, or more than 10 times the revenue analysts expect for the next year, according to LSEG. The multiple is 25% higher than a year ago. Its partly attributable to the prospect of upselling Microsoft 365 subscriptions with AI features and the 30% growth in its Azure web services division, helped along by OpenAI and its peers.

Other top lines are set to be boosted by AI ventures at sums equivalent to the amount of capital being pumped into them. Amazon (AMZN.O), for example, unveiled plans in September to plow up to $4 billion into Anthropic, which has agreed to spend just as much over the next five years on Amazons cloud computing services, according to the Wall Street Journal. And soon after Amazons deal, Google committed an additional $2 billion to the same startup, which months earlier had signed a contract with Google Cloud worth more than $3 billion.

All this mutual back-scratching is reminiscent of the bubble from a generation ago. Then, new entrants as well as more established companies borrowed more than $1 trillion to install telecommunications cables across ocean floors and beyond. Equipment suppliers including Cisco Systems (CSCO.O), Lucent Technologies and Nortel Technologies enjoyed fast top-line growth, and valuations. The problem was that the revenue underpinning the boom wasnt all it was cracked up to be.

The worst form was known as roundtripping. Company A would strike a network capacity deal with Company B, often at the end of a quarter, and then Company B would buy some from Company A at a similar price. Executives from Global Crossing, for example, paid large sums to settle lawsuits related to the practice. Many in the industry were forced to restate their financial results and ultimately went bankrupt.

Legal, but misguided, vendor financing was another problem at the time. Telecom equipment makers provided credit to startups so they could buy the gear they needed. The rationale was that companies like Nortel could put their blue-chip balance sheets to good use. Customers would pay higher interest rates to compensate for the risk, and lock themselves into using a particular suppliers switches and other gizmos. In a business where size matters, competitors that didnt lend put themselves in danger of being left behind.

When the bubble burst, it created a double whammy. Earlier equipment sales went bad, causing defaults and impairing balance sheets, while new orders also dried up, leading to unexpectedly large losses. Network gear manufacturers endured a decline of more than 90% in their collective market capitalization over a decade.

This begs the question of whether history is threatening to repeat itself today with what could be described as vendor equity. Although the latest arrangements are helping puff up valuations on both sides of the transactions, the dangers appear to be more manageable. For companies valued in the trillions, a few billion dollars is a pittance. Much of the invested capital should be returned relatively quickly as AI firms buy back-end services. The industrys mortality rate, like that of so many other fledgling tech ideas, presumably will be high, but if the overall market keeps growing there might even be benefits while riding the wrong horse.

Amazon, for example, seems motivated by the chance to beef up its nascent chipmaking business. The e-commerce giant agreed to sell semiconductors to Anthropic to power its AI-model training efforts. Nvidia (NVDA.O) has at least 80% and as much as 95% of the AI market, according to estimates by analysts. By using Anthropic as a training ground to iterate and improve on its own less-popular alternatives, Amazon might be able to eat into that dominance, a benefit it could reap regardless of Anthropics fate. If Amazon captured, say, 10% of Nvidias share, its chips business alone theoretically would be worth more than $100 billion.

The trouble with investing obsessions is that they do more damage as they grow. In telecom, for example, vendor financing accelerated rapidly into the crash. Optimists feared missing out on a rare opportunity, while cynics realized even bad deals would lift stock valuations. It means that as this latest tech craze grows, it pays to keep a close eye on how much tech incumbents and entrepreneurial AI endeavors prop each other up.

Follow @AnitaRamaswamy on X

Follow @rob_cyran on X

Editing by Jeffrey Goldfarb, Sharon Lam, Aditya Sriwatsav and Streisand Neto

Our Standards: The Thomson Reuters Trust Principles.

Opinions expressed are those of the author. They do not reflect the views of Reuters News, which, under the Trust Principles, is committed to integrity, independence, and freedom from bias.

Read more here:

Breakingviews - AI investments carry whiff of vicious circles past - Reuters