Killing It with Robots

Killing with Robots

While the world is stuck with chatbots and large language models (LLMs), Google DeepMind is building a league of its own. The company is hell bent on building robots and are acing it.

The company has unveiled RoboCat, a self-improving robotic agent, which is able to learn and perform a variety of tasks with different arms, and also generates new training data for self-improvement. It can pick up new tasks with less than 100 demonstrations, reducing the need for human-supervised training.

RoboCat is built on Gato (Spanish for “cat”), a generalist agent for processing language, images, and actions. The model was released in September 2022 with the sole intention of moving beyond text outputs, and becoming a multimodal and multi-embodiment generalist. After first training with Gato, the researchers launched RoboCat into “self-improvement” training with tasks never seen before.

The training process includes:

  1. Obtaining a substantial number of demonstrations, ranging from 100 to 1000, for a new task or robot by utilising a human-controlled robotic arm.
  2. Employing the collected demonstrations to fine-tune RoboCat specifically for this new task or arm, resulting in the development of a specialised spin-off agent.
  3. Enabling the spin-off agent to further improve its performance on the new task or arm by practising it approximately 10,000 times, generating additional training data in the process.
  4. Integrating both the demonstration data and the self-generated data into RoboCat’s existing training dataset to enhance its overall training data.
  5. Utilising the updated training dataset to train a new version of RoboCat, incorporating the newly acquired knowledge and experience from the specialised spin-off agent.

Google DeepMind has clearly shown that it wants to move out of the traditional language processing models and step towards building agents that can do tasks. It is one of the few companies trying to make AI useful, instead of just building chatbots like others.

Lastly, another work from our lab @GoogleAI, this time without any foundation models: just super fast, agile, state-of-the-art control using modern and classical methods 🦾
3. Agile catching: https://t.co/1UU15NAEmY pic.twitter.com/cn8m5hxjpz

— Karol Hausman (@hausman_k) June 20, 2023

In another innovation discussed in a research paper named Agile Catching with Whole-Body MPC and Blackbox Policy Learning, the company shows a robot catching objects thrown at high-speed. The best part about this research is that it does not use any foundational models such as language modelling to achieve the task. Just with simple tracking and interception, the robot catches the balls thrown at it.

Google previewed its vision and capabilities last year with PaLM-E. This embodied a multimodal language model for performing tasks in the real world based on vision and images. Then with RT-1, Google Research used transformers for real-world control. This shows that Google bringing DeepMind back into its ecosystem was indeed a match made in heaven.

Standing Apart

The only close competitor to Google DeepMind in robotics is Boston Dynamics. The company has been upping the game in robotics ever since they released Spot and now the humanoid ‘Atlas’ is also in the making. This is not to say that no other company is building the robotic dream. Elon Musk’s Tesla unveiled Optimus last year, but is still under work and there is no sign of it anymore. Far less optimistic.

OpenAI once had a robotics division which built a robotic arm that could solve the Rubik’s cube. But the company shut it down in 2021. Now, it has decided to bet on it again and invested in a Norway-based startup called 1x.

In 2021, when OpenAI shut its robotics division, Google DeepMind took a huge step towards building more general robots. In a research blog, the company introduced vision-based robotic manipulation based on RGB-Stacking for enabling robots to understand the world and the objects around it.

Microsoft on the other hand is still in the ChatGPT hangover. In February, the company extended its capabilities in robotics arms, drones, and other home assistant robots calling the research ‘ChatGPT for Robotics’.

Interestingly, the company has a robotics lab called AI Lab Projects, where it is experimenting with AI and robots at the same time to automate a lot of tasks. For this, the lab has Paul-E, a 7 degree motion collaborative robot with embedded vision and high-res force control. Still, the research is nowhere close to how much DeepMind invests in the field.

The debate whether embodiment is required for AGI or not goes on forever, possibly, with Google DeepMind’s deep research into the field with integration of language models into machines.

The post Killing It with Robots appeared first on Analytics India Magazine.

Microsoft Adds “On Your Data” Function to Azure OpenAI Service

In a bid to help enterprise customers analyze data quickly, Microsoft recently announced that it has added an “On Your Data” function to Azure OpenAI Service that allows users to read their favorite documents and chat with ChatGPT.

Currently, in preview mode, enterprise customers can instantly develop a company-exclusive AI agent that responds with a huge amount of internal knowledge, or a bot that can interact with users based on their company’s data.

Click here to know more about Azure OpenAI on your data.

Microsoftが自社の好きなドキュメントを読み込ませてChatGPTに答えさせられる「On Your Data」機能をAzure OpenAI Serviceに追加!
これで社内の膨大な知見を食わせて回答してくれる会社専属のAIエージェントの開発や、自社データに基づいてユーザーと対話できるボットの開発が一瞬でできる! pic.twitter.com/lykIsXZHrY

— KAJI | 梶谷健人 (@kajikent) June 21, 2023

Microsoft said that Azure OpenAI On Your Data enables users to run supported chat models such as ChatGPT and GPAT-4 on their data without needing to train or fine-tune models. Further, it said that running models on their data enables them to chat on top of and analyze their data with greater accuracy and speed.

With this, enterprise customers can now unlock valuable insights that help them make better business decisions, alongside identifying trends and partners, and optimize their operations.

Microsoft said it offers customized content of conversational AI. “Because the model has access to, and can reference specific sources to support its responses, answers are not only based on its pre-trained knowledge but also on the latest information available in the designated data source,” said that team, saying that this grounding data also helps the model avoid generating responses based on outdated or incorrect information.

Last week, MachineHack GenAI unveiled DataLyze, a generative AI data analysis tool that lets users analyze data on the go and has made it accessible to everyone. The platform looks to make the process of delving into datasets as effortless as engaging in a conversation. You can try it out here.

The post Microsoft Adds “On Your Data” Function to Azure OpenAI Service appeared first on Analytics India Magazine.

Encina Boosts Speed of Complex Simulations by 506X on Azure

Encina Boosts Speed of Complex Simulations by 506X on Azure Sponsored Content by Microsoft/NVIDIA June 21, 2023

Manufacturers are starting to embrace the concept of a circular economy, which seeks to make products that are more durable and easier to reuse and recycle. It can be anything from an automaker creating components in a way that makes them easier to disassemble and reuse again, to breaking down old products and waste materials into elements that can be used to make new sustainable products.

It is this latter area where Encina, a company whose mission is to “catalyze the transition to a sustainable, circular economy” is focusing its efforts. It produces circular chemicals from end-of-life plastics, which can be hard to dispose of, and single-use plastics, such as those used in packaging, electronics, car parts, and more. To accomplish this, Encina partnered with CPFD Software to design next-generation recycling facilities.

Simulations critical to scaling up operations

Encina uses proprietary technology on plastic waste. “These difficult-to-recycle plastics are our focus, trying to get them into a circular system,” said Sheida Sahandy, Chief Sustainability Officer at Encina. “We’re giving these molecules a kind of infinite life—they can come in as waste and go back out to become almost anything.”

To do this, Encina breaks down long hydrocarbon chains into shorter hydrocarbon components or monomers using catalytic processes. The end products of these processes are circular base chemicals Benzene, Toluene, Xylene, and Propylene (BTX/P). Encina's customers use these chemicals to produce manufactured circular and renewable products of their own.

The work requires a physical facility with large reactors to run the complex chemical process of breaking down plastic waste. Encina has already proven its technology at a process demonstration unit in Texas, and the company is in the process of constructing its first commercial-scale circular manufacturing facility in Pennsylvania.

Enter CPFD Software. Its Barracuda Virtual Reactor simulates 3D, transient behavior in fluid-particle systems such as those encountered in chemical reactions. CPFD's clients use the solution to optimize processes and confidently scale-up operations. The software has become the industry standard tool for troubleshooting, scaling, and reducing the environmental impact of existing chemical processes while accelerating the research, development, and deployment of newer sustainability and circular economy initiatives.

Enter Microsoft Azure high-performance computing (HPC) and accelerated computing from NVIDIA. Designing Encina’s hardware using the CPFD Barracuda Virtual Reactor requires enormous computational capacity. The companies ran one of the largest real-world Virtual Reactor simulations ever on a public cloud platform, taking advantage of an advanced system from Azure and full-stack accelerated computing from NVIDIA.

“We were early adopters of parallelization on NVIDIA technology,” says Peter Blaser, Vice President of Operations at CPFD Software. CPFD makes use of NVIDIA GPUs for accelerated computing and includes multiple-GPU parallelization capabilities enabling Barracuda Virtual Reactor to scale and run on HPC systems, such as NVIDIA DGX™ hardware systems and Azure GPU-accelerated virtual machines (VMs) with multiple GPUs.

The results of the simulations and key points include:

  • The project used an ND A100 v4-series VM with eight NVIDIA A100 Tensor Core GPUs available.
  • The NVIDIA Ampere generation of GPU cards running Virtual Reactor simulations was about three times faster than previous generations of V100 GPUs. Running on a four-GPU VM, simulations were completed 400 times faster than on an on-premises CPU-based workstation.
  • To demonstrate the power of GPU speedups, CPFD revisited a 2012 simulation running on CPUs that modeled 400,000 cells and 6 million computational particles, which would have taken three years to complete and was never finished. Today, a comparable model running on Azure HPC can be completed in 13 hours.
  • In addition to exponentially faster model completion, these speedups allow running larger domain models—models with finer spatial resolution, more detailed physics, and more complex chemical reactions.
  • CPFD and Azure achieved a 506X speed boost running a large-scale model on a single Azure VM compared to the CPU version of the results.

Putting this compute power into perspective, a comparable company-built system would cost more than $100,000. An additional benefit of using these cloud capabilities is that Azure and NVIDIA constantly upgrade their technologies. Thus, Encina always has access to the most performant technology available at any given time; technology purpose-built for HPC and AI workloads.

Bringing it full circle

The aim of Encina’s work is to move beyond traditional mechanical recycling, which is currently the dominant recycling technique for plastic waste. Mechanical processes are only effective for a few types of plastic, so their use is limited. And its use tends to degrade the quality and color of the plastic, so the plastic can only be recycled a few times.

In contrast, Encina’s processing can be applied to a wider range of plastics. That makes it more effective in reducing waste that now ends up in the wrong places. In addition, the process does not cause the same degradation. So, the materials can be used repeatedly to make new products that meet high quality, purity and color standards.

Building systems to do the processing and scaling up operations are complex tasks. Working with CPFD and using Azure’s infrastructure powered by NVIDIA GPUs, Encina can run simulations that allow the company to optimize reactor design and mitigate risks. That, in turn, lets Encina participate in the circular economy.

Learn more

Find out more about Encina.
Find out more about CFPD.
Learn more about Azure HPC and AI Infrastructure.
Learn more about NVIDIA helping power a sustainable future.

#AzureHPCAI #MakeAIYourReality

Related

EU AI draft law: Could Europe’s new AI rulebook set a global precedent?

Ai robot with a law scale and gavel judge beside him.
Image: phonlamaiphoto/Adobe Stock

European lawmakers moved closer to passing a pioneering law on artificial intelligence last week, advancing legislation that aims to set a benchmark for the rapidly evolving — yet minimally regulated — technology.

On Wednesday, June 14, the European Parliament approved draft legislation known as the AI Act. Billed as the world’s first comprehensive AI law, the legislation represents a rulebook for the adoption and use of AI technology in the European Union’s 27 member states.

The AI Act proposes a ban on high-risk AI practices in Europe, including the use of real-time facial recognition technology in public places and other AI systems deemed “intrusive and discriminatory” by the European Parliament, such as social scoring systems and models that employ “subliminal or purposefully manipulative techniques.”

The draft legislation also includes stricter requirements for generative AI models like ChatGPT, which will be forced to disclose when content has been machine-generated and design its models with built-in measures to prevent the generation of illegal content.

Jump to:

  • A precedent in AI law
  • A risk-based approach to AI rules
  • Mixed reactions from the tech community
  • When will the new law pass?

A precedent in AI law

Europe’s pioneering legislation intends to set a precedent for artificial intelligence regulation around the world, where an explosion in the use of AI and machine learning tools has left policymakers scrambling to keep up.

Deirdre Clune, a Member of the European Parliament, heralded the AI Act as “a ground-breaking piece of legislation” with the potential to become “the de facto global approach to regulating AI.”

Speaking to MEPs on June 13, Clune said: “It is among the first global attempts to regulate AI … AI has the capacity to solve the most pressing issues, including climate change or serious illness, and we want to lay the foundations for doing this here in the European Union.”

Clune added: “We cannot do this entirely on our own, but we should be leaders in ensuring that this technology is developed and used in a responsible ethical manner, while also supporting innovation and economic growth.”

A risk-based approach to AI rules

The EU’s draft legislation proposes a risk-based approach to AI regulation that categorizes artificial intelligence systems based on their potential threat to users, which is a topic that has long been the subject of fierce debate.

AI systems deemed to be carrying an “unacceptable” risk level will be strictly prohibited under the EU’s law with limited exceptions. AI systems and functions deemed unacceptable — and therefore banned — under the draft bill include:

  • Cognitive behavioral manipulation of people or specific vulnerable groups: for example, voice-activated toys that encourage dangerous behavior in children.
  • Social scoring: classifying people based on behavior, socioeconomic status or personal characteristics.
  • Real-time and remote biometric identification systems: facial recognition tools are a possible example.

An exception in the case of remote biometric identification systems would be for prosecuting serious crimes in instances where identification occurs after “a significant delay,” though such cases will require court approval.

High-risk AI systems include those used in EU-regulated products like toys and cars, as well as specific areas such as biometric identification, critical infrastructure management, employment and law enforcement. Under new EU rules, these systems must be registered in an EU database.

SEE: Artificial Intelligence Ethics Policy (TechRepublic Premium)

AI systems with the potential to influence voters in political campaigns, as well as those found in recommendation systems used by social media platforms, also feature on the AI Act’s high-risk list.

Meanwhile, generative AI tools such as ChatGPT and Google Bard, as well as other AI systems deemed a limited risk, will be required to adopt stronger safeguards under the new EU rules. These safeguards include stricter transparency requirements and enabling users to make informed decisions about if and how they interact with AI models.

Users will have to be informed when they are interacting with an AI and must also be given the option to cease or continue using AI applications once they’ve interacted with them.

“The absolute minimum that we need to offer here is transparency,” Clune said. “It must be clear that this content has not been made by humans. And we also go one step further and ask developers of these large models to be more transparent and share their information with providers and how these systems were trained and how they were developed. This should address and alter the environmental sustainability of these systems.”

Mixed reactions from the tech community

While Europe’s AI Act ultimately aims to govern the use of artificial intelligence in a way that balances safety and transparency with innovative potential, members of the tech community have voiced concerns that increased scrutiny and potential penalties for breaching the rules could limit innovation.

Kevin Bocek, vice president of ecosystem and community at cybersecurity company Venafi, argued that the European Parliament was “squarely taking aim at Silicon Valley’s AI innovations” and warned of “a potentially huge impact on U.S. business and their investors.”

“The EU will significantly crimp the current approach to AI of weekly product releases and daily model updates,” Bocek told TechRepublic. “The bloc’s requirements for transparency, certification and safety don’t align with how software and cloud providers innovate at present. This opens a path for European startups and open source to play a larger role in AI than we’re currently seeing today.”

The new EU rules could also make things trickier for companies that operate in Europe but are headquartered elsewhere.

Greg Hanson, group vice president of platform sales for EMEA and Latin America at software development company Informatica, shared with TechRepublic that Europe’s AI Act would require U.S. and non-EU businesses to establish full visibility around the origin of the data on which their AI models were built to ensure compliance.

“For organizations with data crossing international borders — which is most — they will now need to have full visibility of how and where their data is processed to meet different geographical legislation,” Hanson told TechRepublic.

“For example, an organization headquartered in the USA but operating in Europe will need to fully understand the quality of its data and be able to trace it fully through their data supply chain …This means the need for data accuracy, clarity, lineage and governance will intensify.”

SEE: Experts laud GDPR at five year milestone (TechRepublic)

The AI Act includes exemptions to rules for research activities and AI components provided under open-source licenses.

To help ensure businesses can effectively harness AI while protecting citizens’ rights, regulatory sandboxes will be established by public authorities in order to test new AI systems before their deployment.

Meanwhile, citizens will have enhanced rights to file complaints about AI systems and receive explanations for decisions based on high-risk AI systems, with the reformed EU AI Office taking responsibility for monitoring how the rulebook is implemented.

Despite this, Kamales Lardi, digital transformation consultant and author of “The Human Side of Digital Business Transformation,” warned that regulators would “continue to play a catch-up game” unless limitations in the draft act were quickly addressed.

“The Act is taking a traditional regulatory and compliance approach to a dynamic and rapidly changing landscape that is generative AI,” Lardi told TechRepublic.

“The Act does not sufficiently address topics around copyright, even from the perspective that debate and definitions around what is considered copyright boundaries are still in discussion.”

Lardi also noted that implementation of the AI Act would “be a nightmare” given the number of companies currently using AI-based solutions or planning to do so in the near future. “The review of applications and conformity assessment will be a daunting task, and relying on self-assessment will not be sufficient in the long term,” she added.

“Companies may need to make substantial changes to their data collection and management practices to meet the new data privacy standards set by the legislation.”

When will the new law pass?

The EU hopes to finalize the AI Act by the end of 2023, though even if successful in doing so, the new legislation is not expected to come into force for a few years — potentially around 2026.

Regardless of the intricacies that will need to be worked out in the interim, Hanson said he welcomes the introduction of Europe’s landmark AI legislation.

“The EU’s decision to regulate the data feeding AI systems is a smart legislative move,” Hanson stated. “Not only will it protect the potential of a technology that can fuel economic growth, but it will protect the very essence of a business,” he said.

“AI, in particular, puts data accuracy on a knife edge. Incorrect data, which ultimately ends up fueling AI models, will have a negative brand impact. But accurate, trustworthy, timely data will give organizations that much-needed competitive edge and drive organizational growth.”

Person using a laptop computer.

Subscribe to the TechRepublic UK Newsletter

Catch up on the week’s essential technology news, must-read posts, and discussions that would be of interest to IT pros working in the UK and Europe.

Delivered Wednesdays Sign up today

How to use Canva AI to turn any photo into a professional headshot

Canva Magic Eraser feature

It doesn't matter what you are wearing in the photo or what the background is — with the help of AI, you can transform it all.

Whether you need to set up your LinkedIn profile for the first time, add a photo to your Slack account, or simply want to revamp your website, a professional headshot is always a good thing to have.

However, hiring a photographer to take a professional headshot can get expensive. Plus I, like many others, often dislike how I look in those posed pictures where I'm trying to look my best. Instead, I prefer the more natural and candid photos taken by my friends and family.

Also: How to use ChatGPT to write a cover letter (and why you should)

Canva's Magic Edit tool helps circumvent both issues by creating a perfect headshot from an image you already have and like.

It doesn't matter what you are wearing in the photo or what the background is — with the help of AI, you can transform it all.

How to use Canva to create the perfect headshot

Before you can get started with your project, you will need to become a Canva Pro member — though Canva does offer a 30-day free trial.

The Magic Edit tool, as well as other AI-powered design tools, such as Magic Eraser and Magic Write, is limited to Pro users at the moment.

Also: How to use Magic Eraser on the Google Pixel

The membership costs $120 per year for an individual account and includes other perks, such as image editing, stock photos, premium templates, and more.

Once you sign up — or initiate the free trial — and are signed in, you can get started on creating your headshot. Here's how you do it.

Final result:

Disclaimer: Using AI-generated images could lead to copyright violations, so people should be cautious if they're using the images for commercial purposes.

More on AI tools

Artists are rightfully upset that ‘Secret Invasion’ used AI Art for opening credits

Artists are rightfully upset that ‘Secret Invasion’ used AI Art for opening credits Lauren Forristal 7 hours

Marvel’s latest series, “Secret Invasion,” made its debut on Disney+ today, sparking tons of backlash after it was confirmed the intro sequence was AI-generated.

“Secret Invasion” director Ali Selim admitted to Polygon that the opening credits were designed by VFX company Method Studios using artificial intelligence. Selim didn’t say which AI tool was used.

Method Studios was not immediately available to comment to TechCrunch. We will update this story if we hear back.

Selim described the generative AI tool to Polygon, however, confessed that he didn’t “really understand” how it works. He said the team would feed the system “ideas and themes and words,” which would then “Go off and do something. And then we could change it a little bit by using words, and it would change,” Selim added.

Image Credits: Marvel Studios

Artists are increasingly worried that AI creations will not only discredit their work but also threaten their livelihoods. Generative AI takes millions of images made by actual people to create a hodgepodge of art that once took someone hours, days, months or even years to produce. To have a multi-billion-dollar studio use AI tech instead of paying real artists is concerning. Plus, this particular opening sequence will likely be seen by millions of viewers.

Many artists took to Twitter to express their frustrations, including Jeff Simpson, who worked as a visual development concept artist on “Secret Invasion.”

“’Secret Invasion’ intro is AI-generated,” Simpson wrote. “I’m devastated, I believe AI to be unethical, dangerous and designed solely to eliminate artists’ careers. Spent almost half a year working on this show and had a fantastic experience working with the most amazing people I ever met…”

Brian Long, an independent filmmaker, tweeted:

A lot of people will rave about how good #SecretInvasion is but, this show marks the first major invasion of AI on a major studio’s piece of work, it’s utterly horrifying to think about and this is only the start.

Think about all the amazing intros created by REAL HUMANS!!! pic.twitter.com/WLRAyR9kM5

— Brian Long (@BrianLongFilms) June 21, 2023

Some users said that they’re boycotting the new series.

While AI has played a role in other Disney content, such as the de-aging of Harrison Ford in the new “Indiana Jones” movie, this is the first time that a Marvel TV show has used the technology.

Marvel has yet to publicly respond to the criticism.

Cisco Launches New AI Networking Chips

Cisco Systems has launched networking chips for AI supercomputers, entering the competition with Broadcom and Marvell Technology. The new chips, part of Cisco’s SiliconOne series, are currently being tested by five major cloud providers, although their names were not disclosed. These cloud providers are likely to include industry leaders such as Amazon Web Services, Microsoft Azure, and Google Cloud.

The increasing popularity of AI applications, such as ChatGPT powered by graphics processing units (GPUs), has emphasized the importance of fast communication between individual chips. Cisco, a prominent supplier of networking equipment, including ethernet switches, has introduced the latest generation of its ethernet switches called G200 and G202. These switches offer double the performance compared to the previous generation and can connect up to 32,000 GPUs together.

According to Rakesh Chopra, a Cisco fellow and formerly principal engineer, the G200 and G202 chips are expected to be the most powerful networking chips in the market for AI and machine learning workloads. They aim to provide a more power-efficient network with 40% fewer switches and reduced lag.

Cisco’s entry into the AI supercomputer chip market puts it in direct competition with Broadcom, which announced the Jericho3-AI chip in April. The Jericho3-AI chip also supports the connection of up to 32,000 GPU chips.

Overall, Cisco’s networking chips target the growing demand for AI applications and aim to offer improved performance, efficiency, and connectivity for AI supercomputers.

The post Cisco Launches New AI Networking Chips appeared first on Analytics India Magazine.

HPE Offers LLM Option for Supercomputing in the Cloud

HPE Offers LLM Option for Supercomputing in the Cloud June 21, 2023 by Agam Shah

The high-performance computing options are thinning, and HPE is one of the last U.S. companies standing to build supercomputers – a fact indirectly acknowledged by Dan Reed, a professor at the University of Utah, during a keynote at the ISC conference last month.

Now, HPE has a new computing option for running HPC-focused machine learning applications without building and managing on-premise supercomputers: the company is expanding its GreenLake cloud service to include high-performance computing options for large language models.

HPE GreenLake for Large Language Models in the cloud “allows single large-scale AI and HPC jobs to run on hundreds or thousands of CPUs or GPUs at once, which is very, very different than general purpose cloud offerings that run multiple jobs in parallel on a single instance,” said Justin Hotard, executive vice president, and general manager for HPC labs and AI, during a press briefing.

Supercomputers require unique datacenter capabilities for power and cooling, and “until now, supercomputers have not been available on demand in a consumption model,” Hotard said.

The supercomputing option in the cloud is part of a larger announcement of HPE entering the cloud market for AI. HPE GreenLake for Large Language Models provides access to a supercomputer, software stack and services for customers to remotely run machine-learning applications. GreenLake packages computing as a utility service – similar to how electric companies charge for monthly usage. In this case, HPE charges for the AI computing consumed in the cloud and provides the ability to customize software and hardware connections.

HPE is providing access specifically to its Cray XD supercomputers, and the “initial large language models are on Nvidia H100 GPUs,” Hotard said, adding “we will provide further details as we launch additional specific types of instances.” The company announced XD2000 and XD6500 supercomputers for enterprise and AI applications last year. The supercomputers use some of the Cray technologies that are also in exascale systems. The XD supercomputers have the SlingShot interconnect and ClusterStor E1000 storage system.

An ongoing debate in the supercomputing community revolves around the relative safety of air-gapped supercomputing systems, and the risks and performance issues in moving HPC workloads to the cloud, which is happening gradually.  On LLMs, the cloud can be a bottleneck, as it does not provide the throughput or bandwidth of on-prem supercomputers. HPE is studying how the on-premise and cloud models for AI could complement each other.

“We’ve been doing a small development cloud testbed for a number of months now and we’ve got a lot of positive feedback on requirements. I think they have got something that we believe will be addressed,” Hotard said, adding that HPE GreenLake for LLMs in the cloud is a complementary offer to its on-prem supercomputers. Customers that run on-premise AI workloads look for bursting capabilities, and the cloud option can be added to the cluster. The cloud can free up on-prem resources to run more critical AI workloads, and it also adds diversity to the HPC software and computing stack, Hotard said.

Services & software

HPE has a stack of hardware and software services at both levels that will complement each other.  The software stack has features for LLMs to be trustworthy and accurate. A machine-learning data management software allows data to be visible at all times, and integrated, tracked, and audited. These capabilities provide guardrails to generate safe and reliable data, which has been a big concern with LLMs hallucinating and generating unstable responses.

“Our supercomputers leverage the HPE Cray programming environment, which offers developers tools to create or debug and tune code and optimize their applications,” Hotard said.

Users will have access to Luminous, a large-language model from Aleph Alpha that has 13 billion parameters. The LLM is multimodal, meaning it can process images and text. Top AI companies such as Google are building multi-modal LLM that supports all types of input, including voice. “This is the first service that we’re announcing on HPE GreenLake, and we anticipate releasing services in other areas such as climate modeling, drug discovery, financial services, manufacturing, and transportation,” Hotard said. Luminous fits into HPE’s plans to offer foundational models that support HPC workloads, as opposed to generic workloads that are used for general-purpose computing. HPE will initially offer open-source models as well as proprietary models that are available for purchase.

“We’re already in deep discussions on the pharmaceutical side as well, for example. Those are going to vary based on the use cases or what the best partners are to work with,” said Evan Sparks, chief product officer for Artificial Intelligence at HPE, in response to a question from HPCwire.

HPE currently has no plans to offer OpenAI’s GPT-4 large language model to supercomputing customers, but “that could evolve as partnership discussions evolve,” Sparks said.

The HPE GreenLake for LLMs will first be publicly available in North America, and then it will be available in Europe towards the end of the year or early next year.

The competition

To be sure, public cloud providers also offer supercomputing options via virtual machine instances. Google last month announced the A3 supercomputer, which has 26,000 Nvidia H100 GPUs. Amazon also offers its own suite of HPC offerings, including EC2 instances with its high-speed Elastic Fabric Adapter interconnect, and the Lustre or ZFS file systems.  Amazon is also talking about confidential computing in its HPC offerings to keep data secure while data is being processed in the VMs.

Customers have multiple options for AI and supercomputing in the cloud, but it will come down to whether organizations want to use HPE’s service, which is a hybrid system and less exposed to the public, or public cloud providers, which have security mechanisms in place to protect data.

But HPE said the GreenLake for LLMs is complementary to public cloud services, where customers may store structured and unstructured data that is used to train machine learning models.

“Data is a critical input to training and tuning these kinds of models. There needs to be a [mechanism] to get data from the public cloud, or wherever the data resides, into the service. Those mechanisms will be made available,” Sparks said. GreenLake customers can also make API calls, which is a popular way to connect AI applications to well-known LLMs such as GPT-4.

Related

Otter.ai’s new AI chatbot will automate your next steps after meetings

otter-ai-chat-banner

Otter.ai began as an AI transcription service and has expanded into enterprise solutions that help make meeting interactions easier with features such as audio-recording meetings, generating notes, transcribing, and generating summaries.

Now it is adding a new major AI feature to help even more — Otter AI chat.

Also: These are my 5 favorite AI tools for work

Users can invite Otter Chat to their meetings where it will act as a meeting participant. Once in the meeting, Otter Chat will be able to provide users with immediate feedback on questions and generate meeting-specific content.

For example, Otter Chat can provide answers regarding discussion points or key decisions that were discussed during a meeting and clarification about details, according to the release.

Otter.ai CEO and co-founder Sam Liang claims that Otter Chat's ability to analyze meetings and provide valuable insight into them is what separates it from AI chatbot competitors such as ChatGPT.

Also: Even Google is warning its employees about AI chatbot use

"Though ChatGPT has great knowledge based on public data, it can't answer any questions about your own team meetings. Also, it's only a 1:1 human-robot chat with no team collaboration," said Liang.

After a meeting, Otter AI can also generate content regarding the meeting, such as a list of action items, summaries, follow-up emails, and blog posts, according to the press release.

That feature could be especially useful for post-meeting deliverables, such as composing emails, making task lists, and more.

Also: Gmail will help you write your emails now: How to access Google's new AI tool

To address potential concerns about having AI sit in on all of your meetings, which can include confidential information, Otter.ai states, "Data used in Otter AI Chat will not be stored by third parties."

The feature will roll out to users within the coming days.

Artificial Intelligence

A thorny question: Who owns code, images, and narratives generated by AI?

Abstract data intelligence through squares outlined in blue

Developers and business people need to better understand the legal implications of generative AI, a new study from Stanford University urges. That's because a wave of court cases may be coming, and the code, language, and images generated through AI may potentially be based on copyrighted material.

The question is, who gets the credit for generative AI output? The study suggests this is still a hazy area.

Also: Who owns the code? If ChatGPT's AI helps write your app, does it still belong to you?

Generative AI enables developers and business users to generate code and narratives at the push of a button. The bad news is that "most of the words and images in datasets behind artificial intelligent agents like ChatGPT and DALL-E are copyrighted," the paper's authors point out. "Existing foundation models are trained on copyrighted material. Deploying these models can pose both legal and ethical risks when data creators fail to receive appropriate attribution or compensation."

The legal doctrine that provides cover for the use of limited segments of the code, narratives, and images is "fair use." However, the study's authors assert, "If the model produces output that is similar to copyrighted data, particularly in scenarios that affect the market of that data, fair use may no longer apply to the output of the model. Fair use is not guaranteed, and additional work may be necessary to keep model development and deployment squarely in the realm of fair use."

Also: Is humanity really doomed? Consider AI's Achilles heel

AI and machine learning practitioners "aren't necessarily aware of the nuances of fair use," Peter Henderson, a co-author of the paper, pointed out in a related interview. "At the same time, the courts have ruled that certain high-profile real-world examples are not protected fair use, yet those very same examples look like things AI is putting out. There's uncertainty about how lawsuits will come out in this area."

The co-authors predict an upcoming wave of lawsuits stemming from uncompensated use of code and content with generative AI. If courts eventually rule that AI does not meet the criteria of fair use, it "could dramatically curtail how generative AI is trained and used. As AI tools continue to advance in capabilities and scale, they challenge the traditional understanding of fair use, which has been well-defined for news reporting, art, teaching, and more. New AI tools — both their capability and scale — complicate this definition."

Also: Serving Generative AI just got a lot easier with OctoML's OctoAI

The IP and copyright implications for code also need to be weighed. "Like in natural language text cases, in software cases, literal infringement — verbatim copying — is unlikely to be fair use when it comprises a large portion of the code base," the co-authors explain. "And when the amount copied is small, the overall product is sufficiently different from the original one, or the code is sufficiently transformative, then fair use may be indicated under current standards."

Such standards may be more permissive than those for text or music. "Functional aspects of code are not protected by copyright, meaning that copying larger segments of code verbatim might be allowed in cases where the same level of similarity would not be permissible for text or music. Nonetheless, for software generated by foundation models, the more the generated content can be transformed from the original structure, sequence, and organization, the better."

Also: If you use AI-generated code, what's your liability exposure?

The report's co-authors recommend establishing technical guardrails. "Install fair use filters that try to determine when the generated work — a chapter in the style of J.K. Rowling, for instance, or a song reminiscent of Taylor Swift — is a little too much like the original and begins to infringe on fair use."

Artificial Intelligence