Andrew Ng’s latest course ‘Generative AI for Everyone’ has set a new record by enrolling close to 43K users in little less than a week. He posted on X, stating that this has been the fastest growing course this year on Coursera, which was launched in the first week of November.
Just got the data: Generative AI for Everyone is the fasting growing course on Coursera this year, with 43K enrollments in first 7 days. Thanks everyone who's signed up! Come join us: Learn about how GenAI works and how to use it! https://t.co/FaENtuU74j https://t.co/Gu8fxcmpnJ
— Andrew Ng (@AndrewYNg) November 13, 2023
The course gives an overview on all aspects of Generative AI and offers students a holistic understanding of this new technology. Presented by Andrew Ng himself, the course is beginner friendly. However, he explained the more technical concepts to better understand how the models work like pre training, fine-tuning or RAG.
This short 6 hour course can be taken over three weeks and covers basic principles and introduces what Generative AI is. It explains the practical applications like tips for prompt engineering, what to keep in mind while choosing a model, and the new workflows possible with Generative AI. Andrew Ng also illustrates how to use the Generative AI tools for the best results like for writing or reading. Like all Coursera courses, this comes with assessments, tests and certification.
Apart from this Andrew Ng, offers comprehensive courses in generative AI. These include topics like diffusion models, GANs, and VAEs, with collaboration from AWS and OpenAI. The courses are free and time-efficient, typically taking one to two hours to complete.
Read More : Top 8 Free Generative AI Courses 2023
The post Andrew Ng’s “Generative AI for Everyone” Enrolls 43K Users in Less than A Week appeared first on Analytics India Magazine.
In a significant development NVIDIA’s RAPIDS AI has recently announced a groundbreaking update to its cuDF library. The new feature, dubbed the ‘pandas accelerator mode’ (cudf.pandas), will change data processing in Python. This mode is now readily accessible online, available on platforms supporting Python GPU DataFrame libraries, such as Google Colab.
The pandas accelerator dramatically enhances the speed and efficiency of data processing by leveraging the robust capabilities of GPU acceleration. The primary aim of introducing this feature is to provide a seamless solution for boosting the performance of existing pandas workflows, all without the need for any alterations in the existing codebase.
Pandas is a powerful data analysis and manipulation library for Python! NVIDIA just made Pandas 150x faster with zero code changes All you have to add is just a couple of lines of code: %load_ext cudf.pandas import pandas as pd Thread pic.twitter.com/donHqUHpgS
— Sumanth (@Sumanth_077) November 13, 2023
With the power of GPUs, the pandas accelerator mode enables a significant reduction in data processing times, which is a crucial factor in handling large datasets and complex computations.
The mechanism behind this acceleration is both innovative and user-friendly. When the pandas accelerator mode is activated using the `%load_ext cudf.pandas` command in a Python environment, it replaces standard Pandas types like Series and DataFrame with proxy objects.
These proxies are designed to direct operations to cuDF wherever feasible, allowing the GPU to efficiently handle the more computationally demanding tasks. This not only ensures a smoother and faster data processing experience but also maintains the familiarity and ease of use of the pandas API.
The recent update to NVIDIA’s cuDF, part of the RAPIDS suite, introduces the ‘pandas accelerator mode’, allowing pandas code to run on GPUs for enhanced performance. This feature was illustrated in a Jupyter notebook tutorial analysing the “Parking Violations Issued – Fiscal Year 2022” dataset from NYC Open Data, demonstrating faster data processing in tasks like grouping and sorting.
Key aspects of this update include its ease of integration into existing pandas workflows, requiring no code modification to activate GPU acceleration. The tutorial also highlights profiling tools within `cudf.pandas` for performance analysis and better resource utilisation understanding.
Furthermore, the update’s compatibility with third-party libraries, such as Plotly Express for data visualisation, showcases its practical application. This enhancement in cuDF is set to notably improve efficiency in data science and analytics.
The post NVIDIA RAPIDS AI Rolls Out Game-Changing ‘cuDF Pandas Accelerator Mode appeared first on Analytics India Magazine.
From OpenAI, to Cohere, and Anthropic, and big giants like Microsoft Azure, AWS, and IBM Watsonx.ai, and the open source LangChain, everyone loves to RAG. So, what’s the deal with RAG, and why is it gaining popularity so fast within enterprise?
RAG, or Retrieval-Augmented Generation, burst onto the scene in 2020 when the brainiacs over at Meta AI decided to jazz up the world of LLMs. It’s a game-changer. Designed to give LLMs much-needed information techniques, RAG swooped in to fix the problems that haunted its predecessors – the dreaded hallucinations.
LLMs rely on statistical patterns without true comprehension. They’re excellent at generating text but struggle with logical reasoning, resulting in hallucinations.
That is because LLMs, no matter how big the size of the model is, and how long the context length is, it is still fixed to that information that it was fed during training. With RAG, customers can add another data set, and give the LLM fresh information to generate the answer from. This is what enterprises need, to generate insights from their own data.
The safety issues
With the launch of GPT-4 Turbo and the Retrieval API, OpenAI has tried to fix the hallucination problem. With the long context length and the option for enterprises to integrate new data for information, OpenAI has almost cracked and solved the most important problem of LLMs, but forgot the data privacy of users.
For example, with a little fancier prompt engineering, a user on X was able to download the original knowledge files from someone else’s GPTs, an app built with the recently released GPT Builder, exactly with RAG. This is a big security issue for this model.
Oh man — you can just download the knowledge files (RAG) from GPTs. I don't know if this is a security leak or "just" a prompt engineering? @OpenAI @simonw https://t.co/VKMW8s4vfb pic.twitter.com/S1RYREna9b
— Kanat Bekt (@kanateven) November 9, 2023
If you give access to your documents to the AI model, someone can “convince” it to let them download the original files. Interestingly, Sam Altman at DevDay made no such announcement about this. Though the release blog conveniently says, “As with the rest of the platform, data and files passed to the OpenAI API are never used to train our models and developers can delete the data when they see fit.”
It seems as if the announcement of GPT Builder was just one more step for OpenAI to collect more data from the users, as long as they don’t delete it. Now that the company is also training GPT-5, it might make use of the files people upload and train on it. If it is just a bug, OpenAI should fix it immediately and make the original file inaccessible to the end user.
Similarly, Google Bard also faced a similar prompt injection problem, where a hacker was able to exfiltrate files such as Docs, Drive, and YouTube history, from the chatbot that other users have uploaded. Even Google’s Bard is not foolproof.
Users on Reddit discuss if LangChain’s RAG offering would be better than using OpenAI’s. Currently, GPT Builder has a 20-file limit on its platform for building a single GPT, which makes it less desirable for serious developers.
Everyone RAG differently
If you can ignore these security flaws with GPTs, it is still a viable user. But all of this should happen in a single prompt which should ask the question and also ask the LLM to retrieve information from the specific dataset. And each company is focusing on solving a specific problem at the moment.
For dynamic knowledge control, RAG lets you tweak and expand its internal knowledge without the hassle of retraining the entire model. This is mostly provided by open source companies such as LangChain, by integrating them with a vector database such as Pinecone, and using it with any open source LLM.
Today, I learned how to code my own Retrieval Augmented Generation (RAG) I just took me a couple of hours. Without RAG, I received an hallucination. With RAG the response is short but correct.#aws #bedrock #llm #ec2mac #python pic.twitter.com/sgRZxQPj8B
— Sébastien Stormacq (@sebsto) November 7, 2023
Every LLM builder does this by trying to expand the size of the model, or in the case of Anthropic, Bard, or Cohere, they let the user get the answers from the internet. This also allows them to generate current and reliable information for not relying on outdated facts. RAG ensures the LLM always has the latest and most trustworthy information at its tips.
For ensuring domain specific knowledge, Cohere and Anthropic let’s enterprise provide their own personal data through Oracle Cloud for expanding on the internal data. These LLMs with RAG, provide insights that are more personalised company’s data.
This in the end definitely brings to question the announcement of the retrieval API by OpenAI. Though the price of it is decreased, other alternatives along with open source ones, make OpenAI’s closed door one unscalable. Though OpenAI is trying to introduce Long-Context RAG with an increased number of tokens, in the hope that users wouldn’t want internet connection.
GPT-4-turbo can now process 128K input tokens! This is the next generation of RAG: Long-Context RAG Longer context windows have the potential to overcome the limitations of search. Search is frequently evaluated based on recall of the top K results. A smaller value of k (1 to… pic.twitter.com/8nu9OW4naH
— Erika Cardenas (@ecardenas300) November 13, 2023
RAG stands out for its unique blend of benefits and cost-effectiveness. Its advantages include dynamic knowledge control, access to current and reliable information, transparent source verification, effective information leakage mitigation, domain-specific expertise, and low maintenance costs, among others. Choose your pick wisely.
The post When, and When Not to RAG appeared first on Analytics India Magazine.
Silicon Valley startup Cerebras Systems and G42, UAE-based technology holding group, have completed the first phase of the AI Supercomputer Condor Galaxy network, featuring nine interconnected supercomputers with a total capacity of four exaFLOPs and 54 million cores.
This achievement initiates the next phase, Condor Galaxy 2 (CG-2), aiming to expand the network to eight exaFLOPs and 108 million cores.
“G42’s strategic partnership with Cerebras Systems is paving the way toward our joint vision of creating the world’s largest and fastest AI supercomputers,” said Kiril Evtimov, Group CTO, G42.
“We support organizations working with unique, massive, and diverse datasets that require extensive AI training capacity. These use cases span across different sectors including healthcare, energy, investments, and climate studies to name a few.” he added.
“In partnership with G42, we are changing the worldwide inventory of compute and using our combined expertise to advance AI work in a powerful way, to quickly and efficiently train leading LLMs.”said Andrew Feldman, CEO and co-founder, Cerebras Systems.
The successful collaboration has resulted in the launch of models like Jais 13B and 30B bilingual models and the Med42 Clinical LLM. Dr. Andrew Jackson, EVP, chief AI officer at Core42, a G42 company, emphasised the platform’s scale and ease of use, demonstrating accelerated AI solution development.
The Condor Galaxy, optimized for training large AI models, addresses significant compute, dataset, and specialised AI expertise requirements. Cerebras and G42’s plan involves deploying nine interconnected supercomputers with a cumulative compute power of 36 exaFLOPs.
Condor Galaxy 2, optimised for LLMs and Generative AI, will deliver four exaFLOPs of 16-bit AI compute, supporting models with up to 600 billion parameters.
The post G42, Cerebras Unveils Brand New AI Supercomputer appeared first on Analytics India Magazine.
After releasing GPT-4 Turbo, OpenAI is silently working on developing GPT-5. Sam Altman mentioned that the training process for GPT-5 will necessitate an increased volume of data, as per FT report. Altman explained that this data would be sourced from a blend of publicly accessible datasets on the internet and exclusive datasets from private companies.
While GPT-5 is likely to be more sophisticated than its predecessors, Altman said it was technically hard to predict exactly what new capabilities and skills the model might have.
“Until we train that model, it’s like a fun guessing game for us,” he said. “We’re trying to get better at it, because I think it’s important from a safety perspective to predict the capabilities. But I can’t tell you exactly what it’s going to do that GPT-4 didn’t.” said Sam Altman, as reported.
Meanwhile, OpenAI also intends to seek additional financial support from its major investor Microsoft to achieve its vision of AGI, as stated by Sam Altman. At the first-ever developers’ conference – DevDay 2023 – OpenAI Chief Sam Altman was super optimistic about building AGI together with Microsoft.
“I think we have the best partnership in tech. I’m excited for us to build AGI together.” said Altman, sharing the stage with Microsoft chief Satya Nadella.
Microsoft invested $10 billion in OpenAI earlier this year as part of a “multi-year” agreement, putting a valuation of $29 billion on the San Francisco-based company, as per sources familiar with the discussions.
OpenAI recently introduced ‘OpenAI Data Partnerships,’ inviting organisations to collaborate in producing both public and private datasets for training AI models. The initiative aims to enhance AI’s understanding of various subjects, industries, cultures, and languages, facilitating the development of AGI, as stated by OpenAI.
The post OpenAI is Silently Working on GPT-5 appeared first on Analytics India Magazine.
In the last two years, I published 5 machine learning and AI books, including one on synthetic data by Elsevier. This represents over 800 pages of compact, state-of-the-art material. The new addition features my most recent advances: the problems that I encountered with generative adversarial networks, and how I overcome them with new techniques. The direction is towards less less training data, yet better results and increased speed. Thus, significant cost savings, especially in the context of synthetic data. Better evaluation metrics and new loss functions with very fast implementation, contribute to the success.
Highlights
This book covers optimization techniques pertaining to machine learning and generative AI. The emphasis is on producing better synthetic data with faster methods, some not even involving neural networks. I describe NoGAN for tabular data is in detail, with full Python code. It includes several case studies in healthcare, insurance, cybersecurity, education, and telecom. This low-cost technique is a game changer: it runs 1000x faster than generative adversarial networks (GAN) while consistently producing better results. Also, it leads to replicable results and auto-tuning. However, I also discuss how to fix failing GANs without abandoning deep learning, and make them replicable.
Better Evaluation Metrics
Many evaluation metrics fail to detect defects in synthesized data. Not because they are bad, but because of partial implementation. Due to the complexity, the full multivariate version of these metrics is absent from vendor solutions. In this book, I describe how to do it for the multivariate Kolmogorov-Smirnov distance (KS). Illustrations cover numerous examples. KS relies on on the joint empirical distributions attached to the datasets. This evaluation metric also works in any dimension on categorical and numerical features. The book features open-source Python libraries, both for NoGAN and KS.
Better Performance With or Without Deep Learning
Then, I discuss a very different synthesizer, namely NoGAN2. It relies on resampling, model-free hierarchical methods, auto-tuning, and explainable AI. Interestingly, it minimizes a particular loss function, also without gradient descent. While not based on neural networks, it nevertheless shares many similarities with GAN. Thus you can use it as a sandbox to quickly test various features and hyperparameters before adding the ones that work best, to GAN. Even though NoGAN and NoGAN2 don’t use traditional optimization, gradient descent is the topic of the first chapter. Applied to data rather than math functions, there is no assumption of differentiability, no learning parameter, and essentially no math. The second chapter introduces a generic class of regression methods. It covers all existing ones and more, whether your data has a response or not, for supervised or unsupervised learning. I use gradient descent in this case.
One chapter focuses on NLP, discussing an efficient technique to process large amounts of text data: hidden decision trees, presenting some similarities with XGBoost. Indeed, I use a similar technique in NoGAN. Then I discuss other GenAI methods and various optimization techniques. In particular, feature clustering, data thinning, smart grid search and more. Finally, I also explain how to use exact multivariate interpolation to synthesize time series and geospatial data. There is also a chapter on agent-based modeling to synthesize to complex system.
Target Audience
Methods are accompanied by enterprise-grade Python code, also available on GitHub. Chapters are mostly independent from each other, allowing you to read in random order. The style is very compact, and suitable to business professionals with little time. Jargon and arcane theories are absent, replaced by simple English to facilitate the reading by non-experts, and to help you discover topics usually made inaccessible to beginners. While state-of-the-art research is presented in all chapters, the prerequisites to read this book are minimal: an analytic professional background, or a first course in calculus and linear algebra.
Getting Your Copy
All my books are available as PDF documents on my e-Store, here. Currently, the only one available in print – “Synthetic Data and GenAI” – is published by Elsevier. The new 200-page book “Statistical Optimization for GenAI and ML” was published in October 2023. The table of contents is on GitHub, here. I am currently working on my next book, “Practical AI & Machine Learning Projects and Datasets”.
Author
Vincent Granville is a pioneering GenAI scientist and machine learning expert, co-founder of Data Science Central (acquired by TechTarget in 2020), Chief AI Scientist at MLTechniques.com, former VC-funded executive, author and patent owner – one related to LLM. Vincent’s past corporate experience includes Visa, Wells Fargo, eBay, NBC, Microsoft, and CNET.
Vincent is also a former post-doc at Cambridge University, and the National Institute of Statistical Sciences (NISS). He published in Journal of Number Theory, Journal of the Royal Statistical Society (Series B), and IEEE Transactions on Pattern Analysis and Machine Intelligence. He is the author of multiple books, including “Synthetic Data and Generative AI” (Elsevier, 2024). Vincent lives in Washington state, and enjoys doing research on stochastic processes, dynamical systems, experimental math and probabilistic number theory. He recently launched a GenAI certification program, offering state-of-the-art, enterprise grade projects to participants.
Financial giant Bloomberg has introduced a new service called Intra-Firm Chatbots to enhance clients’ digital transformation efforts. This service allows Bloomberg Terminal users to incorporate proprietary chatbots into IB chat rooms exclusively for members of the same firm. Intra-firm chatbots assist in surfacing crucial information from internal systems within IB, aiding in-house business intelligence discoverability.
IB Connect, a suite of services, facilitates seamless integration between Bloomberg Terminal and firms’ in-house workflow tools, thereby improving collaboration among colleagues.
The Intra-Firm Chatbots service offers two-way integration, linking a client’s applicable IB chat rooms with their internal systems. It employs natural language processing to organise unstructured IB data and makes the enriched information available to the client’s Intra-Firm Chatbots. Clients can use a provided software development kit to customise these chatbots according to their firm’s tech stack and internal workflow.
Bloomberg offers two types of client chatbot capabilities via IB Connect: Q&A Intra-Firm Chatbots and Notification Intra-Firm Chatbots. Q&A Chatbots enable two-way communication to fetch actionable intelligence from clients’ systems, while Notification Chatbots provide timely alerts and business intelligence within the IB environment without disrupting ongoing team communication.
Back in April, Bloomberg had unveiled BloombergGPT, a specialised LLM tailored for the finance industry. With training based on 700 billion tokens from both Bloomberg’s vast financial data archives and public datasets, this 50-billion parameter model is designed to enhance existing NLP tasks like sentiment analysis, news classification, and query-related tasks in the financial domain.
However, it has limitations, including being monolingual (English only) and potentially carrying biases and toxicity, issues common in LLMs. Unlike its multilingual counterparts like BLOOM and GPT-3, BloombergGPT’s singular language focus could limit its training input diversity.
Read more: What BloombergGPT Brings to the Finance Table
The post After BloombergGPT, Comes Intra Firm Chatbots, Improving Digital Transformation appeared first on Analytics India Magazine.
This article almost didn't get written. I spent most of the weekend falling down a rabbit hole. It was a good rabbit hole. It was a fascinating rabbit hole. But it was not conducive to writing an article.
Also: I took this free AI course for developers in one weekend and highly recommend it
This is what happens when I'm given an all-you-can-eat buffet of really cool and really interesting material to learn and watch. Learning spikes. Productivity tanks.
I'm talking about Amazon's large array of free AI-related courses. After I wrote my article about free OpenAI and DeepLearning courses last month, Amazon approached me to say that it, too, has a library of free AI training.
I was pointed to ten resources it offers. Some, like the Generative AI Foundations playlist, contain multiple learning opportunities. In fact, it's the Generative AI Foundations series that occupied most of my time this weekend.
This deep dive delivers seven hours of video, with topics ranging from how to prepare for creating a foundation model to prompt engineering, prompt tuning, pre-training a new model, preparing data and training at scale, and how to incorporate human feedback into the model and deploy your own foundational model.
Also: How to create your own custom chatbots using ChatGPT
It was good. I spent most of the time with a small, warm, cozy dog on my lap. No writing got done. And that's why I'm here now, late at night before this piece is due, scrambling to get this into my editors' queue before the sun comes up.
Amazon has course offerings for technical and non-technical audiences. For most of these, you'll need to sign into Amazon's AWS Skill Builder site. You can use your AWS account, or even sign in using your consumer Amazon account. That's what I did.
Let's take a look.
Free AI courses for non-technical learners
Unlike the DeepLearning stuff I previously spotlighted which was aimed primarily at programmers, Amazon provides a great overview for managers, decision-makers, and the AI-curious.
Introduction to Artificial Intelligence: This is a brief introductory course that explains what AI is, why it's important, and how it relates to machine learning and deep learning.
Introduction to Machine Learning — Art of the Possible: This is a quick, helpful course that explains the basics of machine learning and how to help evaluate the benefits and risks associated with adopting ML in various business cases. It will be an hour or so well spent.
Generative AI for Executives: This is a collection of free, brief, and easy-to-follow videos to help C-suite executives understand how generative AI can help address their business challenges and drive business growth. Each course is only a couple of minutes, and if you're willing to give up less than half an hour of your day, you'll come out knowing a lot more about what all the fuss is about.
Beginner AI resources for developers and other techies
These resources are suitable for folks starting out with AI. You'll need some quality tech and coding skills to understand what's going on here, but you can get a lot out of it even if you're completely new to AI.
AWS DeepRacer: This isn't a course as much as it's a hands-on learning lab. It's also free, but only to a point. Once you sign up, you get 10 hours and up to 30 days to use those 10 hours. After that, you'll need to sign up for a subscription. That said, you'll be able to squeeze a lot of learning from those 10 hours. DeepRacer lets you go hands-on with machine learning through a cloud-based 3D racing simulator, a fully autonomous 1/18th scale race car driven by reinforcement learning, and a global racing league. It's cool.
Amazon's Machine Learning University: This is a big win for learning. Amazon offers anyone, anywhere the opportunity to access the same machine learning courses used to train Amazon's own developers on machine learning. MLU provides a free, comprehensive self-service pathway to understanding the foundations of machine learning.
Amazon CodeWhisperer – Getting Started: This is a free, self-paced digital course introducing learners to Amazon CodeWhisperer, an AI coding companion designed to help developers get more done, faster. Learners are taught its capabilities, how to set it up, and how to begin using it in their programming language of choice. I didn't know too much about CodeWhisperer before I took this course, but I found out (a) it works with PHP and PhpStorm, my development language and IDE, and (b) it's free for individual use. This has all the signs of becoming a future rabbit hole.
Amazon Bedrock — Getting Started: This is a free self-paced digital course introducing learners to Amazon's service for building generative AI applications. This one-hour course will introduce developers and technical audiences to Amazon Bedrock's benefits, features, use cases, and technical concepts. Bedrock was once an Amazon-only internal tool that's now available to anyone, so it's definitely worth spending the hour learning what it can do.
Intermediate AI resources for developers and other techies
Now, we move on to what Amazon considers intermediate-level courses. There's a lot here, so let's dig in.
Twitch Series: AWS Power Hour Introduction to Machine Learning for Developers: This is a recording of a Twitch-based learning chat series. It helps you learn the foundations of machine learning and get a practical perspective on what developers really need to know to get started with machine learning. It also teaches how you can leverage the power of machine learning and deep learning to make your applications more intelligent.
Generative AI Foundations on AWS: This is the course I describe above, the one that ate my Saturday. It's a free, on-demand technical deep-dive course designed for technologists already familiar with AI modeling. The course includes conceptual fundamentals, practical advice, and hands-on guidance to pre-train, fine-tune, and deploy state-of-the-art foundation models on AWS and beyond
Generative AI with Large Language Models: Now we come full circle. This is a hands-on course that AWS jointly developed with DeepLearning.AI and Andrew Ng, a pioneer in machine learning and education. Those are the folks who created the course I spotlighted in my last free AI learning roundup. This is a three-week course that prepares data scientists and engineers to become experts in selecting, training, fine-tuning, and deploying large language models (LLMs) for real-world applications.
Go ahead. Learn something — for free!
While Amazon and other providers do offer a wealth of paid learning programs, if you dig a little (or just read my articles), you can find a huge amount of free resources to get you started and up your game.
Just remember to ask Alexa to set a couple of alarms or timers before you hit Enroll and start learning. Otherwise, you may lose track of time (in a good way).
Also: How I made this great-looking table in Excel
So what about you? Have you taken these or other Amazon courses? Did you use any of the learning resources I spotlighted last month? Let us know in the comments below.
You can follow my day-to-day project updates on social media. Be sure to subscribe to my weekly update newsletter on Substack, and follow me on Twitter at @DavidGewirtz, on Facebook at Facebook.com/DavidGewirtz, on Instagram at Instagram.com/DavidGewirtz, and on YouTube at YouTube.com/DavidGewirtzTV.
In May, Google announced its Search Generative Experience (SGE), which infuses AI into its market-dominating search engine. Since then, the SGE has been open to user testing via Search Labs, and if you haven't joined the testing pool yet, these latest updates might make you want to.
Although SGE users have already been able to take advantage of helpful AI-powered features such as AI-generated information snapshots in search results and tailored product recommendations while shopping, Google is now adding more features, including an easier way to ask follow-up questions, AI-powered translation help, and broader availability.
Through its SGE testing, Google shares that it has learned that people like to learn more about topics through follow-up questions. As a result, Google is experimenting with a new, easier way to ask follow-up questions from the search results page.
All users have to do is enter their follow-up questions in the "ask a follow-up" fields that are sprinkled throughout the Google search results page, as seen in the video below.
This update is rolling out to US users in English in the upcoming weeks.
Generative AI will also be used to make the translation process more efficient and seamless, as words can have more than one meaning, causing ambiguity in the accuracy of a translation.
Now, with generative AI, users no longer have to guess what the translation actually means. If a word has more than one possible meaning, it will be underlined. Users can then tap on the word and indicate the specific meaning they want to say.
Google shares this feature will be coming soon to the US for English-to-Spanish translations, with the addition of more countries and languages coming in the near future.
In addition to more functional translation features, to expand its accessibility, Google made SGE available in four new languages, including Spanish, Portuguese, Korean, and Indonesian. The company launched SGE in more than 120 new countries and territories around the world.
Lastly, SGE will be able to provide users with interactive definitions on AI-powered overviews for coding and health information, meaning that on relevant searches, users will see certain words highlighted, which they can hover over to get a full definition.
This update is rolling out over the next month for US English users, with more countries and languages being added soon.
Cerebras founder Andrew Feldman on the packing crates for Condor Galaxy 1, or, "CG-1," a massively parallel AI computer built for client G42.
The world is starved for GPU chips from the dominant artificial intelligence vendor, Nvidia. That has so far not produced a meaningful surge in chip sales by competitors Advanced Micro Devices and Intel. But it may be helping to build a new kind of computing model.
"It's increasingly the case that there's, sort of, one alternative to Nvidia," said Andrew Feldman, co-founder and CEO of AI computing startup Cerebras Systems, which sells a massive AI computer, the CS-2, running the world's largest chip.
Also: Nvidia boosts its 'superchip' Grace-Hopper with faster memory for AI
Feldman and team began selling computers to compete with Nvidia's GPUs four years ago. A funny thing happened on the way to market. Feldman is increasingly finding his business is a hybrid one, where there are some sales of individual systems, but much larger sales of massively parallel systems that Cerebras builds over months, then runs on behalf of clients as a dedicated AI cloud computing service.
The business "has changed completely" for Cerebras, Feldman told ZDNET. "Rather than buying one or two machines, and putting a [computing] job on one machine for a week, customers would rather have it on 16 machines for a few hours" as a cloud service model.
The result for Cerebras, is, "For hardware sales, you can do fewer, bigger deals, and you're gonna spend a lot of time and effort in the management of your own cloud."
On Monday, at a supercomputing conference in Denver called SC23, Feldman and team unveiled the latest achievement of that expanding AI cloud.
Also: Cerebras just built a gargantuan computer system with 27 million AI 'cores'
The company announced they have completed the construction of a massive AI computer, Condor Galaxy 1, or, "CG-1," built for client G42, a five-year-old investment firm based in Abu Dhabi, the United Arab Emirates.
"We now have so much super-compute capacity that other people are using our system in all sorts of creative ways," said Feldman of Condor Galaxy.
Condor Galaxy, announced earlier this year, is named for a spiral galaxy located 212 million light years from Earth. The machine is a collection of 64 of Cerebras's CS-2s. The total value of CG-1 is equivalent, said Feldman, to a little bit less than the cost of an equivalent number of Nvidia's GPU chips, on the order of $150 million, based on the price of Nvidia's 8-way "DGX" computer.
"This is a very good business," said Feldman of such large-ticket sales. "We're having a monster year" in terms of sales, he said.
Also: Why Nvidia is teaching robots to twirl pens and how generative AI is helping
The Condor Galaxy machine is not physically in Abu Dhabi, but rather installed at the facilities of Santa Clara, California-based Colovore, a hosting provider that competes in the market for cloud services with the likes of Equinix.
Cerebras is beginning construction of the second version of Condor Galaxy, number two, or, "CG-2," which will add another 64 computers and four more "exaFLOPS" of computing power. (An exaFLOP is a billion, billion floating point operations per second, see Wikipedia), for a combined total of 8 exaFLOPs for the Condor Galaxy system.
The Condor Galaxy system is expected to total, in its final configuration, 36 exaFLOPs, using 576 CS-2 computers, overseen by 654,000 AMD CPU cores.
In the new hybrid business model, said Feldman, the measure of success becomes not just system sales but also the rate of new customers renting capacity in the Cerebras cloud without any up-front purchases. "Sometimes you'd ship them hardware and they'd set it up, and you'd run a trial or prove it out on their premise, and now we give you a login," explained Feldman of the new mode of sales.
Pharmaceutical giant GlaxoSmithKline, an early customer for CS-2 hardware, is also renting capacity in Cerebras's cloud, said Feldman. "They have our gear on-premise, and then, when they want to do giant runs, they come into our cloud," he explained. "And that's a very interesting model."
"We now have, sort of, so much super-compute capacity that other people are using our system in all sorts of creative ways," said Feldman. "In the AI space, they're training interesting models, and in the super-compute space, they're doing interesting work — and this just isn't the case with anybody else."
Feldman cited as "unbelievably interesting" AI work done on Condor Galaxy the development of an open-source large language model, akin to OpenAI's GPT. That program is the best-performing model with 3 billion neural network "parameters" on the machine learning repository Hugging Face, noted Feldman, with more than a billion downloads. That program is small enough to be run on a smartphone to perform AI inference, which is the intention, said Feldman.
As an example of scientific work, Feldman cited a research paper by scholars at King Abdullah University of Science and Technology in Saudi Arabia that was a finalist for the distinguished Gordon Bell Award given out by the Association for Computing Machinery, organizer of the SC23 event.
"We lent them time on Condor Galaxy so they could break records for seismic processing," noted Feldman.
The first version of Condor Galaxy, CG-1, took 70 days to complete, said Feldman. The CG-2 machine will be finished "early next year." The company is already planning Condor Galaxy-3, which will add another 64 machines and another 4 exaFLOPS, for a system total of 12 exaFLOPS.
One of the key advantages of a machine such as Condor Galaxy, both 1 and 2, said Feldman, is the systems' engineering. Putting together an equivalent number of GPU chips is incredibly difficult, he told ZDNET. "The number of people that can network a thousand GPUs is very small," said Feldman. "It's, maybe, 25 companies."
Also: Qualcomm's Snapdragon X Elite brings more AI power to your next PC
"It is very hard to get efficient use of that much distributed compute, it's a very, very difficult problem," said Feldman. "That's one of the problems we fundamentally solve."
Every CS-2 computer in Condor Galaxy 1 and 2 contains one of Cerebras's novel AI chips, the "Wafer-Scale-Engine," or WSE. Those chips, the biggest in the world, each contain 850,000 individual "cores" to process AI instructions in parallel, making them the equivalent of multiple GPU chips.
In addition, the CS-2 computers are supplemented by Cerebras's special-purpose "fabric" switch, the Swarm-X, and its dedicated memory hub, the Memory-X, which are used to cluster together the CS-2s.
pic.twitter.com/donHqUHpgS
(@Sumanth_077) November 13, 2023
Stormacq 
(@sebsto) November 7, 2023
