Vertiv, a global provider of data centres, digital infrastructure and continuity solutions, has launched its AI Hub to address the growing need for expert information on AI infrastructure deployment and strategy. The hub features the industry’s first AI reference design portfolio for critical digital infrastructure.
The Vertiv AI Hub provides partners, customers, and website visitors with access to white papers, industry research, tools, and power and cooling portfolios for retrofit and greenfield applications. The reference design library showcases scalable liquid cooling and power infrastructure supporting current and future chip sets from 10 to 140kW per rack.
“Vertiv has a history of sharing new-to-world technology and insights for the data centre industry,” said Vertiv CEO Giordano (Gio) Albertazzi. “We are committed to providing deep knowledge, the broadest portfolio, and expert guidance to enable our customers to be among the first to deploy energy-efficient AI power and cooling infrastructure for current and future deployments.”
The AI Hub will be frequently updated with new content, including an AI Infrastructure certification program for Vertiv partners, reflecting the rapid changes in the AI tech stack and supporting infrastructure.
Vertiv also recently inaugurated a new manufacturing facility in Chakan, Pune, to meet the surging demand for data centers and supporting infrastructure solutions in India. This facility complements its existing manufacturing facilities in Ambernath and Pune.
The new Chakan facility manufactures thermal management products and solutions tailored for data centers, telecom, commercial, and industrial applications, catering to both domestic and international markets.
Albertazzi has noted that India’s emergence as a data center hub in the APAC region is a key reason for building the third manufacturing facility in the country. In the current landscape of the data center market, Mumbai emerges as the largest hub, while Pune stands out as a rapidly growing contender.
The post Vertiv Launches AI Hub to Bridge Knowledge Gap in AI Infrastructure Deployment appeared first on AIM.
On Tuesday May 14th, Google announced its sixth-generation TPU (tensor processing unit) called Trillium.
The chip, essentially a TPU v6, is the company’s latest weapon in the AI battle with GPU maker Nvidia and cloud providers Microsoft and Amazon, which have their own AI chips.
The TPU v6 will succeed the TPUv5 chips, which came in two flavors: TPUv5e and TPUv5p. The company said the Trillium chip is “the most performant and most energy-efficient TPU to date.“
(Source: Google)
The Trillium chip will run the AI models that will succeed the current Gemini large-language model, Google said at its IO conference in Mountain View, California.
Performance
Google made all-around improvements to the chip. The chip provides 4.7 times more peak compute performance per chip. It also doubles the high-bandwidth memory, internal bandwidth, and chip-to-chip interconnect speed.
“We got to the 4.7x number by comparing the peak compute performance per chip (bf16) of Trillium TPU vs Cloud TPU v5e,” a Google spokeswoman said in an email to HPCwire.
The BF16 performance on TPU v5e was 197 teraflops, and a 4.7x improvement would put BF16 peak performance on Trillium at 925.9 teraflops.
A large performance improvement in Google’s TPUs was long overdue. The TPU v5e’s 197 teraflops BF16 performance actually declined from 275 teraflops on the TPU v4.
Memory and Bandwidth
Trillium chips have next-generation HBM memory but didn’t specify whether it was HBM3 or HBM3e, which Nvidia uses in its H200 and Blackwell GPUs.
The HBM2 capacity on TPU v5e was 16GB, so Trillium will have 32GB of capacity, which is available in both HBM3 and HBM3e. HBM3e provides the most bandwidth.
Up to 256 Trillium chips can be paired in server pods, and inter-chip communication has improved twofold compared to TPU v5e. Google didn’t share inter-chip communication speeds, but they could be 3,200 Gbps, which is two times that of 1,600 Gbps with TPU v5e.
The Trillium TPUs are also 67% more energy-efficient than the TPU v5e, Google said in a blog entry.
Faster Chip Release Cycle
Trillium is replacing the TPU brand name, and will be the branding behind future generations of the chip. Trillium is based on the name of the flower, and not to be confused with AWS’s Trainium, which is an AI training chip.
Google wasted no time releasing its sixth-generation TPU — it hasn’t even been a year since the company released TPU v5 chips.
TPU v4 – introduced in 2020 – hung around for three years until the release of TPU v5. The development of TPU v5 itself was mired in controversy.
Google claimed that AI agents helped floor-plan the TPU v5 chip about six hours faster than human experts.
Researchers connected to the TPU v5 AI design project were fired or left, and the claims are currently under investigation by Nature Magazine. (https://www.hpcwire.com/2023/10/03/googles-controversial-ai-chip-paper-under-scrutiny-again/)
The Systems
Server pods will host 256 Trillium chips, and the AI chips will communicate two times faster than similar TPU v5 pod setups.
The pods can be combined into larger clusters, and communication occurs via optical networking. Communication between pods will also be two times faster, providing the scalability required for larger AI models.
“Trillium TPUs can scale to hundreds of pods, connecting tens of thousands of chips in a building-scale supercomputer interconnected by a multi-petabit-per-second datacenter network,” Google said.
A technology called Multislice strings large AI workloads across thousands of TPUs in a large cluster. That ensures high uptime and power efficiency of TPUs.
The Chip
The chip has third-generation SparseCores, an intermediary chip closer to high-bandwidth memory, where most of the AI crunching takes place.
The SparseCores bring processing closer to the data in the memory, supporting the emerging computing architecture being researched by AMD, Intel, and Qualcomm.
Typically, data has to move from memory to processing units, which consumes bandwidth and creates chokepoints. The sparse computing model tries to free up network bandwidth by moving processing units closer to memory clusters.
“Trillium TPUs make it possible to train the next wave of foundation models faster and serve those models with reduced latency and lower cost,” Google said.
Trillium also has TensorCores for matrix math. The Trillium chip is designed for AI and won’t run scientific applications.
The company recently announced its first CPU, Axion, which will be paired with Trillium.
The Hypercomputer
The Trillium chip will be part of Google’s homegrown Hypercomputer AI supercomputer design, which is optimized for its TPUs.
The design merges compute, network, storage and software to meet varying AI consumption and scheduling models. A “Calendar” system meets hard deadlines on when a task should start, while the “Flex Start” model provides guarantees on when a task will end and deliver results.
The Hypercomputer includes a software stack and other tools to develop, optimize, deploy, and orchestrate AI models for inference and training. This includes JAX, PyTorch/XLA, and Kubernetes.
The Hypercomputer will continue to work with GPU-optimized interconnect technologies, such as the Titanium offload system and technology, which is based on the Nvidia H100 GPUs.
Availability
Expect the Trillium chips to be available in Google Cloud, but Google did not provide an availability date. It will be a top-line offering, costing more than TPU v5 offerings.
The high prices of GPUs in the cloud may make Trillium attractive to customers. Customers already using AI models available in Vertex, which is an AI platform in Google Cloud, may also switch to Trillium.
AWS’s Trainium chip is also available, while Microsoft’s Azure Maia chip is mainly for inference.
Possible Relief From the GPU Squeeze
Google has historically presented its TPUs as an AI alternative to Nvidia’s GPUs. Google has released research papers comparing the performance of TPUs to comparable Nvidia GPUs.
Google recently announced it will host Nvidia’s new GPU, B200, and specialized DGX boxes with Blackwell GPUs.
Nvidia also recently announced it would acquire Run.ai in a deal valued at $700 million. The Run.ai acquisition will allow Nvidia to keep its software stack independent of Google’s stack when running AI models.
The TPUs were initially designed for Google’s homegrown models, but the company is trying to better map to open-source models that include Gemma, an offshoot of Gemini.
Chinese investor and serial entrepreneur Kai Fu Lee recently said that he is bullish about OpenAI becoming a trillion-dollar company in two to three years’ time.
“OpenAI will likely be a trillion-dollar company in the not-too-distant future (two to three years),” said Kai Fu Lee at a recent event with Fortune.
“I am very bullish on OpenAI’s future. They’ve really done an admirable and unbelievable job executing. Even today, GPT-4 is still the gold standard. You see, Gemini Ultra and Claude 3 make these claims, but if you use these models, GPT-4 and GPT-4 Turbo are unbelievably good and a great balance for, uh, performance and cost,” he added.
Further, he said that despite his concerns about their lack of openness, he greatly admires them. “If I could invest in any one of them (Microsoft, Google and OpenAI), which I can’t, but if I could, it would be OpenAI,” he said.
In March last year, Lee launched 01.AI with the vision of developing a homegrown large language model for the Chinese market. Surprisingly, it has taken the open source route, unlike OpenAI, which is closed source.
OpenAI recently released GPT-4o at its latest Spring Update event, which won hearts with its ‘omni’ capabilities across text, vision, and audio. OpenAI’s demos, which included a real-time translator, a coding assistant, an AI tutor, a friendly companion, a poet, and a singer, soon became the talk of the town.
OpenAI has made GPT-4o available to users for free. “We are a business and will find plenty of things to charge for, and that will help us provide free, outstanding AI service to (hopefully) billions of people,” said Altman.
Altman said that they are yet to figure out ways to make an expensive technology like GPT-4 available to users for free. He emphasised that while they aim to provide advanced AI tools for free or at a minimal cost as part of their mission, the high expenses currently pose a significant barrier.
OpenAI reached the $2 billion revenue milestone in December, according to a report by the Financial Times. The report indicated that OpenAI expects to more than double this figure by 2025, driven by strong interest from business customers looking to implement generative AI tools in the workplace.
The post OpenAI Will Likely Be a Trillion-Dollar Company in Two to Three Years appeared first on AIM.
Arc Browser has become my default — at least on MacOS. (Hello, Browser Company: Please port this app to Linux.) Anyone who's already started using this browser knows that the company behind it is quick to release new features to keep up with — and, in some cases, surpass — the competition.
One such feature is ChatGPT integration. Unlike many other browsers that have incorporated their own take on AI, Arc Browser opts for an already-established AI. However, unlike other browsers, Arc's AI integration isn't quite as obvious. That's OK, however, because once you know how to use it, it's fairly straightforward.
Also: 5 ways Arc browser makes browsing the web fun again
Let me show you how to use ChatGPT in Arc Browser.
First, a cautionary note: When using the Arc Max features (the AI-centric features on Arc Browser), the browser does send data to third parties. For example, when using Ask On Page, the partner is Anthropic. For all other features, the data goes to OpenAI. If you're not OK with that, I would not advise enabling Arc Max.
How to use ChatGPT in Arc Browser on MacOS
What you'll need: The only thing you'll need for this is an updated version of Arc Browser on MacOS (Arc Max is not yet available for Windows — or Linux (see above request). One thing to note: If you want to be able to save your ChatGPT chats, you'll need to sign into OpenAI on Arc Browser before you start using ChatGPT.
Accessing the Arc Max Settings pop-up is done through the command bar.
Also: How to use ChatGPT (and how to access GPT-4o)
If you only want to enable ChatGPT in Arc Browser, here's where you do it.
You now have access to ChatGPT queries.
You can now type your ChatGPT query, which will open a new Arc Browser tab with the response.
Also: I replaced Google Search with Opera's Aria AI and I don't miss the former one bit
No, it's not nearly as easy as, say, Opera's Aria AI solution, but Arc Browser does make using ChatGPT fairly easy. Hopefully, this feature will arrive on the Windows version soon (and, ahem, a Linux port?). If you're an Arc Browser user and a fan of AI, you'll enjoy this ChatGPT integration.
Data engineers historically have toiled away in the virtual basement, doing the dirty work of spinning raw data into something usable by data scientists and analysts. The advent of generative AI is changing the nature of the data engineer’s job, as well as the data she works with–and ETL software developer Matillion is right there in the thick of the change.
Matillion built its ETL/ELT business during the last tectonic shift in the big data industry: the move from on-prem analytics to running big data warehouses in the cloud. It takes expertise and knowledge to extract, transform, and load business data into cloud data warehouses like Amazon Redshift, and the folks at Matillion found ways to automate much of the drudgery through abundant connectors and low-code/no-code interfaces for building data pipelines.
Now we’re 18 months into the generative AI revolution, and the big data industry finds itself once again being rocked by seismic waves. Large language models (LLMs) are giving companies compelling new ways of serving customers when text is the interface and an actionable new data source.
But LLMs and the coterie of tools and techniques that surround them–vector databases, retrieval augmented generation (RAG), prompt engineering–are also enabling companies to do old things in new ways through copilots and autonomous agents. One of the older things that GenAI has targeted for a facelift is ETL/ELT, and Matillion is at the front of that transformation.
Matillion’s AI Strategy
Like many other data tool makers, Matillion has developed an AI strategy for adapting its business and tools to the GenAI revolution.
Copilots help with coding work (Phonlamai Photo/Shutterstock)
On the one hand, the company is updating its existing tools to enable data engineers to work with unstructured data (mostly text) that is the feedstock for GenAI applications. To that end, it’s adapted its software to work with the new data pipelines being built for GenAI applications. That includes connecting into various vector databases and RAG tools, such as LangChain, that developers are using to build GenAI applications, according to Ciaran Dynes, Matillion’s chief product officer.
“There’s a skill in building that. It doesn’t come cheap,” Dynes tells Datanami. “A lot of what we’ll see in Matillion is plain old ETL pipelines–prepping the data, cutting out all the junk, the non-printable characters in PDF, stripping out all the headers and footers. If you send those to an LLM, I’m afraid you’re paying for every single token.”
Matillion is also adopting GenAI technology to improve the workflow in its own products. Earlier this year, the company unveiled Matillion Copilot, which allows data engineers to use natural language commands to transform and prepare data.
The copilot, which will soon be in preview, gives engineers another option for building ETL/ELT pipelines in addition to the low code/no code interface and the drag-and-drop environment.
According to Dynes, the copilot works with Matillion’s Data Pipelining Language, or DPL, to convert natural language requests to transform data using scripts written in SQL, Python, dbt, LangChain, or other languages. In the right hands, Matillion Copilot can enable data analysts to build data transformation pipelines.
“A copilot will definitely help the business analyst be faster, cheaper, better, as well as opposed to needing or always needing the data engineer to fix the data for them,” Dynes said.
Creating AI Pipelines
Matillion developed its ETL/ELT chops working primarily with structured data. But GenAI works predominantly on unstructured data, including text and images, and that changes the nature of the new data pipelines that are being created.
For instance, matching a particular data source into the appropriate table in the destination isn’t always straightforward, as there can be variations in the semantic meanings of data values that machines have a hard time picking up. This is where Matillion has focused much of its energy in creating Copilot.
In Dynes demo, viewer ratings of movies are being loaded into a vector database in preparation for use in a prompt to an LLM. The trouble starts immediately with the word “movies.” What does that mean? Does it include “film”? What about “ratings”? Is that the same as “quality”?
“You can send in information called user context and you can teach a large language model, for the purpose of movie rating, ‘movie’ and ‘film’ are interchangeable words,” Dynes said. “What does quality mean? You look within the database, and maybe it doesn’t have the thing called ‘quality,’ but maybe it has ‘user score.’ To you and me, oh, that’s quality, but how does the how does the machine know the quality and user score interchangeable?”
To alleviate these challenges, Matillion gives users the ability to set rules within Copilot that link certain concepts together. As the user works in the copilot to fine-tune the data that will be used in the prompt, she’s able to see the results in a visual sample at the bottom of the screen. If the data transformation looks good, she can move on to the next thing. If there’s something off, she keeps iterating until it’s right.
Ultimately, Matillion’s goal is to leverage AI to lower the barrier to entry for data transformation work, thereby allowing data analysts to developer their own data pipelines. That will leave data engineers to tackle more difficult tasks, such as building new AI pipelines between unstructured data sources, vector databases, and LLMs.
“The hardest thing is basically teaching the data engineers the new practice called prompt engineering. It is different,” he said. “AI pipelines are not [traditional ETL]. It’s unstructured data, and the way that you work with using this natural language prompt is actually a real skill.”
Hallucinations are a concern. So is the tendency of LLMs to go into “Chatty Kathy” mode. Getting data engineers to prompt the LLMs, which are probabilistic entities, to give them more deterministic output requires some targeted teaching.
“If you do not tell the model to say ‘answer yes or no only,’ it will give you a big blob of text. ‘Well, I don’t know. Do you really like Martin Scorsese movies?’ It will just tell you a lot of bunch of garbage,” Dynes said. “I don’t want to get all that stuff! If I don’t have a yes/no answer or a number, I can’t do analytics on it.”
Matillion Copilot is slated to be released later this year. The company is currently accepting applications to join the preview.
In a discussion on Conversations with Tyler, Peter Thiel, former CEO of PayPal said, “People have told me that they think within the next 3-5 years, the AI models would be able to solve all the US maths olympiad problems,” while highlighting that the future is worse for people in into maths than words.
“The Silicon Valley in the 21st century is way too biassed towards the maths people,” Thiel added, saying that the exact reason for this is not sure. “But that’s the thing that seems deeply unstable and that’s what I would bet on getting worse,” he added.
Maths ability has become the test for everything. Thiel said that people who have to go to medical school have to study maths and calculus and people are weeded out through their proficiency in it. “I am not sure that it’s really correlated with your dexterity in neurosurgery,” he added. “I don’t want someone operating on my brain to be doing prime number factorisation in their head.”
Thiel also said that this is similar to the bias he had for rating people on the basis of their chess skills. But this was later removed in 1997, when IBM’s Deep Blue was able to beat the chess champion Garry Kasparov. “Isn’t that what’s going to happen to maths and isn’t that a long overdue rebalance of our society,” he added.
Similarly, During an episode of the Logan Bartlett Show, Sam Altman recalled how calculators were perceived in his maths classes. “We never got to use calculators,” he said, adding that conversely you had to be proficient with calculators in real-life to excel later.
The importance of learning maths has been emphasised from our school days. It’s a crucial requirement in order to excel as an engineer. Further, with computer science becoming so mainstream, society has started to separate those who are good at maths from those who are not.
Now AI is bringing down this wall for the better. With the advent of tools such as ChatGPT and Copilot, everyone is increasingly becoming a developer without needing to learn maths, democratising access to fields that once required deep knowledge of the subject.
But funnily enough, AI is not yet good at maths, though its capabilities are increasing. Recently, ChatGPT with the Wolfram plug-in scored a 96% in the UK A-level paper for maths, which is an essential qualification to get into the AI field.
What this tells us is that if AI is able to crack an exam that is meant to get into AI, there needs to be a major change in the educational systems across the world to adjust to the shifting paradigm of mathematical teaching.
The post Peter Thiel Says AI is Bad News for People with Maths Skills, Not Writers appeared first on AIM.
Amazon is planning a significant upgrade to its voice assistant, incorporating generative AI technology to help Alexa better compete with advanced chatbots like OpenAI's ChatGPT and Google's Gemini, according to a CNBC report. This upgraded version of Alexa will reportedly not be included as part of the company's $139 yearly Amazon Prime membership. The subscription pricing for the enhanced service has yet to be determined.
Also: ChatGPT vs. Microsoft Copilot vs. Gemini: Which is the best AI chatbot?
In a competitive landscape dominated by tech giants like Apple, Microsoft, and Google, Amazon's announcement of a subscription-based, AI-enabled Alexa will likelyspark great interest among tech analysts and consumers alike. With Apple poised to unveil AI upgrades at WWDC in June 2024, Microsoft infusing its products with Copilot, and Google planning to integrate Gemini across Android and its other offerings, Amazon must demonstrate substantial added value to justify a paid subscription for Alexa.
The future success of Alexa could hinge on its ability to provide unique features and benefits that set it apart from rivals. Indeed, these features should be part of any AI consumer offering — including those from Apple, Microsoft, and Google — if these companies expect consumers to pay for them via ongoing subscriptions.
Here are five ways Amazon (and its competitors) could make an AI subscription worthwhile:
1. Proactive assistance
Anticipating needs: A truly proactive AI assistant could provide information and suggestions without being prompted. For instance, it could remind you to leave for an appointment based on traffic conditions, ensuring you arrive on time. Alexa could integrate with your calendar, monitor real-time traffic data, and understand your typical travel times. Additionally, if you regularly purchase certain items, Alexa could remind you to reorder groceries or suggest booking a maintenance service before it's due.
Daily insights: AI assistants could analyze your habits and provide daily insights tailored to your routine. For example, a morning briefing could include weather updates, traffic conditions, and a summary of your day's schedule. An AI assistant could deliver a comprehensive start-of-day overview by integrating calendars, news sources, and weather services. Additionally, it could offer end-of-day summaries, highlighting your accomplishments and suggesting improvements for the next day.
Also: The newest Echo Show 8 is down to $95 ahead of Memorial Day
Contextual awareness: Developing contextual awareness would significantly enhance Alexa's ability to provide proactive assistance. For example, if you're cooking dinner and you ask Alexa about recipes, the assistant could automatically suggest side dishes, cooking tips, or wine pairings based on what you have in your kitchen and what you're preparing.
Seamless integration with daily activities: Proactive assistance could extend to various aspects of your daily life, from health and fitness to home management. For example, the assistant could provide reminders to take medication, suggest healthy meal options, or offer motivational prompts to keep up with workout routines. For home management, Alexa could remind you of maintenance tasks, like changing air filters or paying bills on time.
2. Advanced customization
Personalized experiences: AI assistants can provide personalized experiences by learning from user interactions and adapting to individual preferences. For example, if a user enjoys listening to jazz in the evening and catching up on the news in the morning, the assistant could suggest doing these activities at the appropriate times.
Behavior analysis: AI assistants can offer tailored recommendations and actions based on understanding user behaviors, habits, and preferences. For instance, if a user frequently requests workout music in the morning, the assistant could automatically start playing the user's favorite playlist when it detects the morning routine.
Also: 3 ways Gemini Advanced beats other AI assistants, according to Google
Personalized routines: Smart home assistants can improve their routines to be more intuitive, anticipatory, and capable of learning from experience. For example, a better "Good Night" routine could involve dimming the lights, setting an alarm for the next morning, and playing relaxing music. Similarly, an improved "Morning Workout" routine could include turning on the lights, playing an energizing playlist, and providing a weather update. This routine would be based on user interactions and preferences, allowing assistants to automate actions based on detected activities, such as adjusting the thermostat and lights when starting a movie.
User profiles: AI assistants can support multiple user profiles within a household, offering personalized experiences for each family member. Each profile could include individual preferences for music, news, reminders, and more.
Advanced notifications: AI assistants can offer advanced notification settings, allowing users to receive alerts and reminders based on their interests and schedules. For example, if a user is a sports fan, the assistant could notify them of upcoming games, scores, and highlights.
3. Services integration
E-commerce, financial data, and meal planning: AI assistants should expand their capabilities by connecting to third-party services like email providers, Instacart, and other e-commerce platforms. This capability would enable them to suggest meal plans based on grocery purchases and dietary preferences. Integration with financial services like Rocket Money and Plaid would allow the assistant to analyze purchases and key personal or business transactions for important insights.
Music and news sources: Integrating music services like Apple Music, Spotify, and multiple internet news sources would allow AI assistants to offer highly personalized music and news recommendations. For example, if you listen to a specific podcast on Spotify every morning, the assistant could start playing it when you wake up.
Health and fitness integration: By integrating with health and fitness apps, AI assistants like Alexa could provide personalized workout recommendations, track your progress, and offer motivational tips.
Also: I tested this cheaper Oura Ring competitor for women — here's my buying advice
Educational and professional tools: Integrating educational platforms and professional development tools would make AI assistants valuable resources for learning and career advancement. For students, the assistant could offer study tips, provide access to online courses, and help manage study schedules. For professionals, the assistant could assist with time management and suggest skill development resources based on career goals.
4. Family and private instances
Private LLM instances: Providing private or family-specific large language model (LLM) instances would enhance privacy and enable more personalized interactions. This could be particularly useful for managing family schedules, sharing reminders, and coordinating activities.
Customized family features: AI assistants could develop family-specific features that integrate seamlessly into family life. Personalized recommendations for each family member would ensure everyone's preferences are considered.
Coordinated family scheduling: While some assistants can integrate with multiple Google, Microsoft, and Apple calendars per user, they cannot perform insights across many users. By accessing everyone's schedules, AI assistants could help coordinate activities, avoid conflicts, and send reminders for upcoming events.
Enhanced privacy controls: AI assistants could provide robust privacy controls to ensure that personal information remains secure. Features could include voice recognition to differentiate between family members, customizable privacy settings for each user, and the ability to review and delete voice recordings.
Also: Microsoft and Khan Academy offer a free AI assistant to all US teachers
Parental controls and safe content: AI assistants could include advanced parental controls that allow parents to restrict the content accessible to their children. They could also offer educational content and interactive learning tools that are both safe and beneficial for children. Additionally, Alexa could be equipped to monitor instant messenger and text platforms to detect signs of harassment or bullying, alerting parents and providing insights into their children's well-being.
Health and wellness support for families: AI assistants could support family health and wellness by integrating with health apps and devices, tracking fitness goals, reminding family members to take medications, and providing health tips tailored to each person's needs.
5. Comprehensive integration and accessibility
Replacing other AI products: Consumer access to Amazon's Titan LLM through Alexa should be compelling enough to replace other AI products, such as OpenAI's ChatGPT and Microsoft's Copilot. To achieve this capability, Titan LLM must offer seamless integration with popular apps, functioning as a plugin across devices.
Ubiquitous access: Titan LLM should be ubiquitous and provide instant access wherever users need it. Whether through a browser extension, a mobile app, or directly integrated into operating systems, users should be able to invoke Titan LLM effortlessly.
Speed and efficiency: To stand out from competitors, Alexa powered by Titan LLM needs to be faster and more efficient. Instant responses to queries and seamless task execution will be crucial.
Seamless integration across platforms: AI assistants should function seamlessly across various platforms and devices, similar to how Grammarly operates as a plugin on browsers and a keyboard plugin on iPhone and Android. This capability would involve developing plugins and integrations for popular productivity tools, social media platforms, and other commonly used apps.
Also: Smart home starter pack: Top 5 devices you need
Deep integration with productivity suites: Any consumer LLM, especially Amazon's, should offer deep integration with major productivity suites like Microsoft Office, Google Workspace, and Apple's iWork. This integration would enable users to perform complex tasks directly through Alexa and Amazon's cloud services.
Cross-platform compatibility: It's important to ensure that consumer-focused AI assistants can work smoothly across various operating systems and devices. Amazon doesn't dominate mobile or desktop operating systems like Google, Apple, and Microsoft, so having its AIs embedded into its products out of the box will be crucial. This cross-platform compatibility will enable users to switch between devices without losing access to their personalized settings and data.
Accessibility for all users: AI assistants should be designed with accessibility in mind, ensuring that all users, including those with disabilities, can benefit from their features. This accessibility could involve integrating with assistive technologies and providing customizable interfaces for users with visual or hearing impairments.
Also: Saving hours of work with AI: How ChatGPT became my virtual assistant for a data project
Enhanced security and privacy: AI assistants must offer robust security and privacy features. Users must trust that their data is secure and that their interactions remain private.
Customizable plugins and extensions: AI assistants should support customizable plugins and extensions that enable users to tailor an assistant to their needs. For example, professionals in different industries could access specialized plugins offering industry-specific insights and tools.
Will consumers pay for AI Alexa or not?
To transition Alexa successfully to a paid model, Amazon must provide upgrades that significantly enhance the user experience and offer unique, valuable features. By focusing on proactive assistance, advanced customization, integration with other services, family and private instances, and comprehensive integration and accessibility, Amazon could make an Alexa subscription worthwhile. These improvements would justify the cost and position Alexa as a leader in the evolving landscape of AI-driven smart assistants, helping it stand out amidst competitive offerings from Apple, Microsoft, and Google.
UAE-based Core42 is building an AI supercomputer with 172 million cores which will become operational later this year.
The system, Condor Galaxy 3, was announced earlier this year and will have 192 nodes with Cerebras WSE-3 chips. The WSE-3 megachip is over 50 times larger than Nvidia’s GPUs and packs significantly more horsepower.
Cerebras WSE2 vs Nvidia GPU (Source Cerebras)
“Deployment of the system, also called CG-3, in Texas will start next month and be completed by September or October,” said Niall Ó Broin, senior director of product management for HPC at Core42, during a presentation at the ISC 2024 supercomputing conference in Hamburg, Germany.
Core42 is emerging as a significant player in datacenters and AI. Last month, Microsoft invested $1.5 billion in G42, Core42’s parent company, to expand the reach of its AI offerings and Azure cloud.
The CG-3 will have 192 nodes of CS-3 servers, which will host the megachips. Each WSE-3 chip has 900,000 cores and 4 trillion transistors, and it is made using the 5-nm process.
“The system will pack a tremendous amount of horsepower. It will have nine square meters of silicon once fully deployed,” Ó Broin said.
Each WSE-3 chip can reach 125 petaflops of peak AI performance. That adds up to 24 exaflops of system performance for the CS-3.
“As part of this chip, there will be four terabytes of on-chip memory. That’s not DRAM or HBM. It’s on-chip SRAM,” Ó Broin continued.
Talking about storage, Ó Broin mentioned, “The system will have 12 petabytes of mass storage. We have VAST across our HPC storage infrastructure.”
Core42 last year acquired Cerebras systems with 64 CS-2s, which it used to develop JAIS, a bilingual Arabic and English language model.
The CS-2 had the WSE-2 chip, which was made using the 7-nm process and had 850,000 cores.
“One thing we really like about Cerebras in these systems is we’ll be able to use our CS-2 code and bring that on to the CS-3 using PyTorch.”
G42 has one system on the Top500 list. Artemis, which has Intel’s 24-core Skylake server CPUs and Nvidia V100, entered the Top500 November 2020 list at 26 and is currently at 129.
Another system, POD3, made by Huawei with Intel chips, exited the list in 2022.
Core42 was formed in 2023 from the merger of G42 Cloud and G42 Inception AI. The parent company, G42, was founded in 2018.
Core42 is also working with hardware from Nvidia and AMD. It is also working with OpenAI and other companies on AI models.
G42 was being scrutinized by the U.S. government as being a conduit through which China would have access to the latest GPUs from Nvidia and other AI hardware from U.S. companies.
Bloomberg reported last month that G42 reached a secret deal with the U.S. government to divest from China so the company continued to have access to Nvidia GPUs.
G42 is a heavy user of Nvidia’s H100 GPUs. Microsoft invested $1.5 billion in G42 around the same time as the secret pact with the U.S. government.
Copilot for Windows is going through a redesign to make Microsoft's AI assistant easier to access and use. Currently available through a sidebar panel, Copilot will soon become a regular resizable and movable Windows app, Microsoft revealed in a blog post on Wednesday.
After the transformation, you'll trigger Copilot through an icon in the middle of the Taskbar. Since it will be a regular Windows app, you'll be able to move, resize, and snap its window to more effectively use it and multitask with other windows on the screen.
Also: Microsoft's latest Windows 11 security features aim to make it 'more secure out of the box'
After you click the icon, Copilot currently appears as a static and unmovable sidebar that's a struggle to navigate and juggle amid open apps and windows. By default, if you click anywhere else on the screen, the sidebar disappears. The only tricks you can perform are to change the width of the sidebar and show it side by side with another window so it remains visible.
"To integrate more seamlessly into everyday workflows and deliver AI-powered assistance in a more convenient manner, we are evolving Copilot in Windows into a standalone application," Microsoft said in its blog post. "With this change, users of Copilot will get the benefits of a traditional app experience such as resizing, snapping, and moving the window."
Consumers using Copilot+ PCs, which Microsoft unveiled earlier this week at its Build conference, will be able to launch Copilot with a single click of the Copilot key on their keyboard. Plus, Copilot for Microsoft 365 subscribers will still be able to maintain separate "web" and "work" tabs to better protect their data.
Microsoft didn't give a specific ETA for the new Copilot for Windows app's availability. In the blog post, the company said only that it's taking a phased and measured approach to this rollout. Indeed, the changes are already popping up in the latest build, Windows 11 Insider 26080, available in the Canary channel.
Also: Microsoft Copilot vs. Copilot Pro: Is the subscription fee worth it?
"We are beginning to roll out an updated Copilot in Windows experience that adds the ability to switch between the existing 'docked' behavior that attaches Copilot to the side of your desktop, and a new mode where it acts like a normal application window which you can resize and move around your screen," Microsoft said in its release notes for the build.
The new build is part of the Windows 11 2024 Update, which is due for general release in H2 2024.
With the help of generative AI, researchers from MIT and the University of Basel in Switzerland have developed a new machine-learning framework that can help uncover new insights about materials science. The findings of the study were published in Physical Review Letters.
As water transitions from liquid to solid state, it undergoes significant transformation properties, such as its volume and density. Phase transitions in water are so common that we don’t even think about them, but this is an intricate physical system. The behavior of materials can be highly complex and challenging to predict at molecular levels during phase transitions.
MIT and University of Basel researchers harnessed the power of GenAI to create a new framework that can automatically map out phase diagrams for novel physical systems, and detect transitions between them.
Scientists have long been puzzled by the abrupt and unpredictable nature of phase changes at the molecular level. The diversity of materials and their properties, combined with sparse scientific data added to the challenge. That is all about to change with the development of this new framework which signals a major leap in the discovery of novel materials and the understanding of their thermodynamic properties.
“If you have a new system with fully unknown properties, how would you choose which observable quantity to study? The hope, at least with data-driven tools, is that you could scan large new systems in an automated way, and it will point you to important changes in the system. This might be a tool in the pipeline of automated scientific discovery of new, exotic properties of phases,” says Frank Schäfer, a postdoc in the Julia Lab at CSAIL and co-author of a paper on this approach.
Schäfer was joined in the research by first author Julian Arnold, a graduate student at the University of Basel; Alan Edelman, applied mathematics professor in the Department of Mathematics and leader of the Julia Lab; and senior author Christoph Bruder, professor in the Department of Physics at the University of Basel.
The research breakthrough makes it possible for scientists to discover unknown phases of matter. The transition of water from liquid to solid is the most obvious example of phase changes. There are other more intricate and more complex material transitions, such as when a material changes conductivity properties with state changes.
(NicoElNino/Shutterstock)
Traditional scientific methods relied on a theoretical understanding of physical states and required scientists to build phase diagrams manually. Such methods had severe limitations, including the inability to map out phase diagrams for highly complex systems, the risk of introducing human bias, and limitations in only theoretical assumptions about which parameters were significant.
The team from MIT and the University of Basel employed physics-informed GenAI models to analyze “order parameters”, a measurable quantity that indicates the degree of order across the phase transition. For example, an order parameter could be used to define what proportion of water molecules are in a structured state versus those that remain in a disordered state.
The Julia Programming Language, known for its outstanding performance in scientific and technical computing, was instrumental in crafting the new ML models. Reportedly, the method published in the paper outperforms other ML techniques in terms of computational efficiency.
The research holds the potential to transform the fields of materials science and quantum physics. Not only can the new framework be useful in solving classification tasks in physical systems, but can also play a key role in improving large language models (LLMs) by identifying how certain parameters can be fine-tuned for better outputs.
Related Items
Google DeepMind’s New AlphaFold Model Poised to Revolutionize Drug Discovery
Drug Developers Enlist Google Cloud HPC in the Fight Against ALS
MIT Researchers Leverage AI to Identify Antibiotic That Can Kill Drug-Resistant Bacteria
