Рубрика: AI

Google Deepmind introduced a successor to its Robotics Transformer model 1 called RT-2, a Transformer-based model trained on text and images from the web, enabling it to directly produce robotic actions.

Unlike chatbots, robots face real-world challenges, requiring a grounding in the physical environment and complex tasks. However, RT-2 is a significant step towards creating more capable and helpful robots, addressing the challenges of time-consuming and expensive training methods used previously. Similar to how language models learn from web data to understand general concepts, RT-2 employs web data to inform and guide robot behaviour.

It is an advancement that extends the capabilities of vision-language models (VLMs), which take images as input and generate text. It builds upon models like PaLI-X and PaLM-E and adapts them to serve as the foundation for RT-2. To enable robot control, RT-2 represents actions as tokens in its output, similar to language tokens, allowing actions to be processed using standard natural language tokenizers. This approach enables the model to output robotic actions and control the behaviour of robots effectively.

Tests and Abilities

Deepmind conducted qualitative and quantitative experiments on RT-2 models using over 6,000 robotic trials. Three categories of skills were defined: symbol understanding, reasoning, and human recognition, which required combining knowledge from web-scale data and the robot’s experience.

RT-2 demonstrated emergent robotic skills that were not present in the robot data, thanks to knowledge transfer from web pre-training. For instance, by leveraging knowledge from a vast web dataset, RT-2 understands concepts like identifying trash and throwing it away, without the need for specific training. It can even grasp abstract concepts, recognizing that certain objects become trash after use.

RT-2 simplifies the process of instructing robots by combining complex reasoning with robotic actions in a single model. It can perform tasks even without explicit training for them. RT-2’s ability to transfer knowledge from language and vision training data to robot actions showcases its versatility and effectiveness in handling various tasks.

It showed more than a 3x improvement in generalization performance compared to previous baselines like RT-1 and VC-1. RT-2 retained performance on original tasks seen in robot data and significantly improved performance on previously unseen scenarios, showcasing the benefits of large-scale pre-training. Moreover, RT-2 outperformed baselines pre-trained on visual-only tasks, indicating its superior performance in handling novel situations.

Google ventured into developing smarter robots by incorporating its language model, LLM PaLM, into robotics, resulting in the PaLM-SayCan system. However, the new robot demonstrated some imperfections during a live demo. The New York Times witnessed the robot inaccurately identifying soda flavours and misidentifying fruit as the colour white.

Others in the Game

While Google DeepMind has been at it when it comes to robotics, Boston Dynamics has also bolstered its efforts and is one of the leading competitors. Boston Dynamics has made significant advancements in robotics with the release of robots like Spot and the improved capabilities of its humanoid robot ‘Atlas.’

Atlas is now capable of navigating uneven terrain, recovering from falls, carrying objects, opening doors, climbing ladders, and performing various tasks. These improvements are a result of enhanced grasping and manipulation capabilities and new control algorithms, allowing Atlas to improvise and adapt to different conditions, at par with top-notch developments, if not more than them.

The robot’s 28 hydraulically operated joints and various sensors, such as LIDAR and cameras, contribute to its flexibility and understanding of its surroundings. Boston Dynamics has a history of developing advanced robots, including Spot and Handle, with the goal of creating versatile robots that can perform a wide range of activities.

While other companies, like Tesla, are attempting to enter the robotics space, they face challenges in making progress, leaving their projects still under development.

OpenAI, on the other hand, had a robotics division that created a robotic arm capable of solving the Rubik’s cube. However, the company shut down this division in 2021. Yet, OpenAI has now decided to re-enter the robotics domain and has invested in a Norwegian startup called 1x.

In 2021, Google DeepMind made strides in building more generalized robots through vision-based robotic manipulation based on RGB-Stacking. This technology enables robots to understand the environment and objects around them.

Meanwhile, Microsoft seems to be focusing on the development of ChatGPT, extending its capabilities to robotics arms, drones, and home assistant robots. The company’s AI Lab Projects division is experimenting with AI and robots together to automate various tasks using the collaborative robot Paul-E, which possesses embedded vision and high-res force control. However, Microsoft’s research efforts in robotics are not as extensive as those of Google DeepMind.

Google DeepMind is deeply involved in researching the integration of language models into machines, which could potentially impact the ongoing debate about embodiment’s significance for AGI.

While Google is focused on Generative AI and AI at large, federated learning still remains one of its main focuses.

Overall, the robotics landscape is highly competitive, with various companies investing in different approaches and technologies to push the boundaries of what robots can achieve.

The post Google Takes Leap Forward in Robotics with RT-2 appeared first on Analytics India Magazine.

OpenAI has been crystal clear about one of the most important aspects behind the success of ChatGPT — Reinforcement Learning from Human Feedback (RLHF). Everyone nodded. And since then, they have all been building models using RLHF.

By training LLMs through interactions with human evaluators, RLHF seeks to improve the performance of AI models in real-world applications, but in turn it induces biases and reduces the robustness of the models. A recent paper, Open Problems and Fundamental Limitations of Reinforcement Learning from Human Feedback by researchers from Harvard, Stanford, MIT, UC Berkeley, and many other universities, discusses the problems with the RLHF approach.

Good, but not the best

According to the paper, obtaining high-quality feedback from human evaluators is one of the primary challenges in RLHF. Human beings, while capable of providing valuable feedback, are susceptible to various limitations and biases. Misaligned evaluators might have difficulty in understanding the context or objectives of the AI model, leading to suboptimal feedback. The complexity of supervision, especially in long conversations, can also hinder the accurate assessment of model performance.

“AI Alignment” is just a cancel culture for GPUs.

— Bojan Tunguz (@tunguz) July 19, 2023

Besides, data quality is another critical concern. Human evaluators may unintentionally provide inconsistent or inaccurate feedback due to factors like limited attention, time constraints, and cognitive biases. Even with well-intentioned evaluators, disagreement can arise due to subjective interpretations and varying perspectives.

The form of feedback used in RLHF can further compound these challenges. Depending on the evaluation method, evaluators may provide binary judgments, rankings, or comparisons, each with its own strengths and weaknesses. Selecting the most appropriate form of feedback for a specific AI task can be complex, leading to potential discrepancies in the training process.

A fundamental issue in RLHF is accurately representing individual human values with a reward function. Human preferences are context-dependent, dynamic, and often influenced by societal and cultural factors. Designing a reward function that encompasses the complexity of human values is a formidable task. Incorrect assumptions about human decision-making or using a reward model that neglects personality and context-dependence can lead to misaligned AI models.

Why so much alignment?

The diversity of human evaluators further complicates the reward modelling process. Different evaluators may have unique preferences, expertise, and cultural backgrounds. Attempting to consolidate their feedback into a single reward model might overlook important disagreements and result in biassed AI models that favour majority opinions. This could be of disadvantage to underrepresented groups and perpetuate existing societal biases.

To address these challenges, researchers must explore techniques for representing preferences in more nuanced and context-aware ways. Utilising ensemble reward models that consider multiple evaluators’ feedback, or personalised reward models that cater to individual preferences, can help capture the diversity of human values.

Transparently addressing potential biases in the data collection process and conducting thorough evaluations to identify and mitigate harmful biases are essential steps in responsible AI development.

To overcome these data constraints, researchers should explore methods for cost-effective data collection that do not compromise data quality and diversity. Understandably, training on GPT-output data for quicker alignment has been the new trend, but this in the end brings in the same bias into other models as well. So, there has been no conclusion on this so far.

The fundamental challenges of RLHF have significant implications for AI alignment. While some problems may have tractable solutions through technical progress, others may not have complete solutions and may require alternative approaches. Researchers must be cautious about relying solely on RLHF for AI alignment, as certain challenges might not be fully addressed through this method alone.

Essentially, RLHF leads to over-finetuning of a model that may handicap its capabilities. This phenomenon is called the alignment tax of AI models. When a model goes to several benchmarks testing with humans in the loop trying to make the model as aligned and as “politically correct” as possible, it loses a lot of its performance.

Alignment tax is the extra cost that an AI system has to incur to stay more aligned, at the cost of building an unaligned or uncensored model, which ultimately also hinders its performance. That is why, in a lot of cases, uncensored models that do not go through the RLHF phase actually perform better than aligned models.

The post AI Alignment is a Joke appeared first on Analytics India Magazine.

There are thousands of articles explaining the differences between data scientist and machine learning engineer. Data science gets broken down even further, with data analysts contrasted to researchers. Professionals skilled in all these domains are called unicorns and believed not to exist. Indeed, they may not work for companies, and ignored when applying for a job. This article explains how to become one, and the benefits that come with it — both for yourself and for employers.

The Silo Mentality

Funneling people into narrow specialized roles is present both in education and in the industry. How many professionals have a degree spanning across ML engineering, statistics, operations research and marketing, to name one of the potential combinations? Such curricula don’t exist. You would have to accumulate multiple degrees. This is cost and time prohibitive. And once in the workforce, you face the same compartmentalization. Sure you may start as a data scientist or business analyst and become MLops engineer, or the other way around. But you can’t be all at once unless you work for multiple employers. Managers and HR can not handle this. They wish such people exist, recognize the benefits. But tunnel vision and long tradition prevents this from happening. Also, HR lacks experience to detect and assess the value of a “unicorn”. Over time, unicorns don’t apply anymore.

Why you Need a Dual Engineer / Scientist?

Companies realized a while back that hiring separate R, SQL and Python programmers is inefficient. Now they want people who know the three languages, or abandon R over time. Likewise, they realize that data scientists are unable to write production code and deal with the full data pipeline. Hiring now favors MLops professionals over scientists. But many ML engineers do not understand well the statistics and business thinking behind the scene. It can result in faulty or inadequate solutions.

Why not hiring someone who masters both? I discuss later how to become one and where to find them (clue: in the job applications that you receive, for instance). Many times, a team of two is linked to the equation 1 + 1 bigger than 2. But hiring a dual engineer / scientist is an example where you can have 1 bigger than 1 + 1. I illustrate this in the next section.

These dual professionals are full stack ML engineers, or full stack analysts. A terminology first coined to define software engineers / web developers that master both back-end and front-end.

When 1 is bigger than 1 + 1

I train software engineers to add ML science to their professional background. Each has his own programming environment, and brings his own datasets to the classroom. My code, which involves complex libraries such as TensorFlow, has to work smoothly on all platforms and datasets. My background is research, but I spend more time designing scalable and generic solutions, than inventing state-of-the-art algorithms. Running SDKs in my cloud, and dealing with incompatible libraries, complex data structures, or metadata, is part of my daily routine. I hire engineers to help on occasion. But because I know both research and engineering, I am able to develop better solutions faster. The integration of engineering and research takes place in one brain (mine), not too. Intricacies in each are resolved jointly via fast human intelligence: all connections and neurons are in a single brain optimized for this dual interaction. Thus 1 is bigger than 1 + 1 here. And less expensive.

Becoming a Dual Engineer / Scientist

It took me many years to become good enough at both. You don’t need hyper-specialization in each role. The trick is to continuously practice the two for long enough. This approach is more efficient than being a scientist for 5 years, than an engineer for 5 years. You could compare it to learning two languages, like English and French: do it simultaneously from day one (at birth) and you will soon outperform — in each language — people who acquired the two languages sequentially. The issue: there are very few corporate jobs or classes offering this opportunity. But you can learn one on the job, and the other one during your leisure time, working on real projects in both. My ML science classes are also a good investment for engineers.

Then you will have more job options. Or if working as a scientist, automate your tasks like an engineer to work very little (or for multiple employers!) If you become an independent consultant or start-up founder, or even a hacker, this dual experience will be invaluable.

Where to Find These Unicorns?

I like to say that unicorns are those who excel in just one job category, not two or three. Most of my connections — executive, founders, consultants — would qualify as unicorns. In my circle, it is the norm. Also, do you really want to hire one? They can scare hiring managers who are reluctant to hire people more competent than they are. Also HR may be unable to identify them due to limited horizon, faulty applicant tracking systems, and inability to connect the dots in a unicorn resume. The rise of generative AI may help with this, leaving the task of finding and hiring unicorns to tools like GPT, trained on LinkedIn profiles and GitHub portfolios.

About the Author

Vincent Granville is a pioneering data scientist and machine learning expert, founder of MLTechniques.com and co-founder of Data Science Central (acquired by TechTarget in 2020), former VC-funded executive, author and patent owner. Vincent’s past corporate experience includes Visa, Wells Fargo, eBay, NBC, Microsoft, CNET, InfoSpace. Vincent is also a former post-doc at Cambridge University, and the National Institute of Statistical Sciences (NISS).

Vincent 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 also the author of “Synthetic Data and Generative AI” (Elsevier), available here. He lives in Washington state, and enjoys doing research on stochastic processes, dynamical systems, experimental math and probabilistic number theory.

Two days after AIM said that it’s time for OpenAI to launch GPT-5, the company filed a trademark application for “GPT-5” with the United States Patent and Trademark Office (USPTO) on July 18. This move suggests the potential development of a new version of their language model. The news was shared by trademark attorney Josh Gerben on Twitter on July 31.

OpenAI has filed a new trademark application for:
"GPT-5"
The filing was made with the USPTO on July 18th.#openai #chatgpt4 #ArtificialIntelligence pic.twitter.com/PhQI3YV3jJ

— Josh Gerben (@JoshGerben) July 31, 2023

The trademark application says that GPT-5 is related to computer software for generating human speech and text, as well as for natural language processing, generation, understanding, and analysis. It is speculated to be the next powerful version of OpenAI’s generative chatbot, following the previous release of GPT-4 in March.

Despite the trademark application, there is no confirmation of immediate development for GPT-5. While it is likely that OpenAI has plans for an advanced language model in the future, the primary purpose of the trademark filing might be to secure the name “GPT-5” and prevent unauthorised use by others.

GPT-5 is anticipated to be the next iteration of OpenAI’s large language model. After the pause letter by the likes of Elon Musk and Steve Wozniak, Sam Altman decided to not train the successor to GPT-4 “for some time,” saying that the company anyway has a lot of work to do before starting to build the model. This was in April. Cut to June, Altman said that OpenAI has not yet started training GPT-5 and is only going to focus on building new ideas. Nevertheless, the specific features and enhancements of GPT-5 have not been officially confirmed by OpenAI.

Moreover, OpenAI had also filed for a trademark on ‘GPT’ with the USPTO in December 2022. OpenAI petitioned in April to the USPTO for hastening the process because a lot of apps named after GPT were springing up.

However, the application is still up and pending and might take up to 4-5 months more to get approved, as Jefferson Scher, a partner in the intellectual property group of Carr & Ferrell told TechCrunch. In a bid to catch up on the delay, the company has released brand guidelines on its website to ensure that no claims are made by people building an AI or GPT disguising as OpenAI.

GPT-4.5 is already here!

There have been claims that the next version of GPT would be superintelligent. Developer Siqi Chen had mentioned that GPT-5 is expected to complete training by the end of the year and could potentially achieve AGI. If GPT-5 does attain AGI, it could significantly boost AI-driven productivity by automating complex cognitive tasks. However, there are concerns and differing opinions, with some experts suggesting that AGI might not be achievable using GPT’s current methods.

In a podcast on Latent Space, Simon Willison, Alex Volkov, Aravind Srinivas, and Alex Graveley argue that Code Interpreter is actually GPT-4.5. It might be just that the company does not want to use the terminology as of now given all the backlash of the pause letter.

In a recent blog about alignment of AI models, OpenAI said that superintelligence might be achieved within four years. The timeline does not match. If the company plans to compete with the rising capabilities of other companies’ AI models, it might ditch the halt on training GPT-5 and actually build it right away.

The post OpenAI Files Trademark for ‘GPT-5’ appeared first on Analytics India Magazine.

Ola’s Aggarwal sets eyes on AI, semiconductor design Manish Singh 7 hours

Indian entrepreneur Bhavish Aggarwal, co-founder of ride-hailing firm Ola and electric vehicle startup Ola Electric, is venturing into fresh terrain as he navigates his businesses towards initial public offerings.

He has set up an AI startup that seeks to develop a large language model and is currently scouting two U.S.-headquartered AI startups for a potential acquisition, people familiar with the matter said. Aggarwal is also in talks to raise over $50 million for the new AI venture, the people said.

The entrepreneur, who founded Ola over a decade ago, has also floated the idea of setting up a semiconductor design firm, one person said, requesting anonymity as the details are not public. It’s unclear whether the semiconductor design firm will be part of the same AI venture.

A spokesperson for Aggarwal declined to comment Monday.

AI and semiconductor designing are the latest of a long-list of areas that Aggarwal has explored in the past decade. Ola leads the Indian ride-hailing market whereas Ola Electric has assumed a leader position in India’s electric scooter market with nearly 250,000 vehicles sold in the past year and a half, according to Society of Manufacturers of Electric Vehicles.

He told Bloomberg last month that Ola had turned profitable whereas Ola Electric had “grown and matured faster” than initial plans, prompting him to advance the timeline for the EV startup’s initial public offering.

The recent surge in AI interest has prompted a boost to the tech economy, delivering a rally in tech stocks and generating a flurry of startup activity. OpenAI’s unveiling of ChatGPT has been a key trigger for the enthusiasm, leading investors to deploy over $20 billion into AI startups in the past quarters. However, India, despite being one of the most significant startup ecosystems, appears to be lagging in this race.

“Normally science enables technologies. AI technology will enable significant acceleration of scientific progress. Science today is still experimental, empirical and relies on the time and creativity of the scientist,” he tweeted last week. “AI will give the scientist significant creative and intelligence leverage. Can we in India become a leading science ecosystem by adopting AI across scientific domains?”

On the other hand, Aggarwal’s assertive expansion into new and often unrelated sectors and their subsequent corporate structures have previously rattled some of his investors. Many backers of Ola, for instance, have expressed concerns about not getting a stake in Ola Electric, which spun out of the ride-hailing firm, people familiar with the matter said.

With the proliferation of generative AI in the business world today, it’s critical that organizations understand where AI applications are drawing their data from and who has access to it.

I spoke with Moe Tanabian, chief product officer at industrial software company Cognite and former Microsoft Azure global vice president, about acquiring trustworthy data, AI hallucinations and the future of AI. The following is a transcript of my interview with Tanabian. The interview has been edited for length and clarity.

Jump to:

• Trustworthy data comes from a mix of human and AI knowledge
• Balancing public and private information is key
• Questions to ask to cut down on AI hallucinations

Trustworthy data comes from a mix of human and AI knowledge

Megan Crouse: Define what trustworthy data is to you and how Cognite sees it.

Moe Tanabian: Data has two dimensions. One is the actual value of the data and the parameter that it represents; for example, the temperature of an asset in a factory. Then, there is also the relational aspect of the data that shows how the source of that temperature sensor is connected to the rest of the other data generators. This value-oriented aspect of data and the relational aspect of that data are both important for quality, trustworthiness, and the history and revision and versioning of the data.

There’s obviously the communication pipeline, and you need to make sure that where the data sources connect to your data platform has enough sense of reliability and security. Make sure the data travels with integrity and the data is protected against malicious intent.

SEE: Major tech players support guidelines for AI safety and cybersecurity, which are similar to recent White House recommendations (TechRepublic)

First, you get the data inside your data platform, then it starts to shape up, and you can now detect and build up the relational aspect of the data.

You obviously need a fairly accurate representation of your physical world in your digital domain, and we do it through Cognite Data Fusion. Artificial intelligence is great at doing 97% of the job, but in the last 3%, there is always something that is not quite there. The AI model wasn’t trained for that 3%, or the data that we used to train for that 3% was not high-quality data. So there is always an audit mechanism in the process. You put a human in the mix, and the human captures those 3%, basically deficiencies: data quality deficiencies [and] data accuracy deficiencies. Then, it becomes a training cycle for the AI engine. Next time, the AI engine will be knowledgeable enough not to make that same mistake.

We let ChatGPT consult a knowledge graph, that digital twin, which we call a flexible data model. And there you bring the rate of hallucinations [down]. So this combination of knowledge that represents the physical world versus a large language model that can take a natural language query and turn it into a computer-understandable query language — the combination of both creates magic.

Balancing public and private information is key

Megan Crouse: What does Cognite have in place in order to control what data the

internal service is being trained on, and what public information can the generative AI access?

Moe Tanabian: The industry is divided on how to handle it. Like in the early days of, I don’t know, Windows or Microsoft DOS or the PC industry, the usage patterns weren’t quite established yet. I think within the next year or so we’re going to land on a stable architecture. But right now, there are two ways to do it.

One is, as I mentioned, to use an internal AI model — we call it a student model — that is trained on customers’ private data and doesn’t leave customers’ premises and cloud tenants. And the big teacher model, which is basically ChatGPT or other LLMs, connects to it through a set of APIs. So this way, the data stays within the customer’s tenancy and doesn’t go out. That’s one architecture that is being practiced right now — Microsoft is a proponent of it. It’s the invention of Microsoft’s student-teacher architecture.

The second way is not to use ChatGPT or publicly hosted LLMs and host your own

LLM, like Llama. Llama 2 was recently announced by Meta. [Llama and Llama 2] are available now open-source [and] for commercial use. That’s a major, major tectonic shift in the industry. It is so big, we have not understood yet the impacts of it, and the reason is that all of a sudden you have a fairly well-trained pre-trained transformer. [Writer’s note: A transformer in this context is a framework for generative Al. GPT stands for generative pre-trained transformer.] And you can host your own LLM as a customer or as a software vendor like us. And this way, you protect customer data. It never leaves and goes to a publicly hosted LLM.

Questions to ask to cut down on AI hallucinations

Megan Crouse: What should tech professionals who are concerned about AI hallucinations have in mind when determining whether to use generative AI products?

Moe Tanabian: The first thing is: How am I representing my physical world, and where is my knowledge?

The second thing is the data that is coming into that knowledge graph: Is that data of high quality? Do I know where the data comes from? The lineage of the data? Is it accurate? Is it timely? There are a lot of dimensions now. A modern data op platform can handle all of these.

And the last one is: Do I have a mechanism that I can interface the generative AI large language model with my data platform, with my digital twin, to avoid hallucinations and data loss?

If the answers to these three questions are clear, I have a pretty good foundation.

Megan Crouse: What are you most excited about in regard to generative AI now?

Moe Tanabian: Generative AI is one of those foundational technologies like how software changed the world. Mark [Andreesen, a partner in the Silicon Valley venture capital firm Andreessen Horowitz] in 2011 said that software is eating the world, and software already ate the world. It took 40 years for software to do this. I think AI is gonna create another paradigm shift in our lives and the way we live and do business within the next five years.

Subscribe to the Innovation Insider Newsletter

Catch up on the latest tech innovations that are changing the world, including IoT, 5G, the latest about phones, security, smart cities, AI, robotics, and more.

Delivered Tuesdays and Fridays Sign up today

After investing $3 billion in AI, Accenture has announced further plans for accelerating generative AI in enterprise. Teaming up with ServiceNow and NVIDIA, the company jointly announced the introduction of AI Lighthouse, a groundbreaking initiative aimed at expediting the development and implementation of enterprise generative AI capabilities.

AI Lighthouse builds upon the existing strategic partnerships among ServiceNow, NVIDIA, and Accenture and will assist pioneering customers across various industries in creating and deploying new generative AI use cases.

By combining the ServiceNow enterprise automation platform and engine, NVIDIA AI supercomputing and software, and Accenture AI transformation services, this comprehensive offering allows customers to collaborate as design partners in architecting custom LLMs and applications, enabling them to drive their businesses forward. This will include self-service options, content generation, and code recommendations.

For this, the partnership will tap into Accenture’s Center for Advanced AI, focusing on generative AI and large language models, to accelerate the design and engineering of domain-specific LLMs and generative AI capabilities within the ServiceNow platform. The aim is to make functional and industry workflows more intelligent, from elevating agent productivity and impact to improving self-service quality and speed with AI-powered virtual agents.

NVIDIA will provide the computing power for model training and tuning with NVIDIA DGX AI supercomputing and NVIDIA DGX Cloud, as well as the NVIDIA NeMo LLM software. ServiceNow will serve as the front-end workflow automation and intelligence platform, while Accenture will leverage its deep functional and industry knowledge to bring these use cases to life for customers.

Speaking on this partnership, Accenture’s Chair and CEO, Julie Sweet, highlighted the enormous potential of generative AI for enterprises, enabling them to reinvent their work processes, enhance services, differentiate themselves, and achieve higher performance levels. She expressed excitement about the partnership with ServiceNow and NVIDIA and their combined experience, expertise, and insights being applied to create powerful and responsible generative AI use cases.

ServiceNow’s Chairman and CEO, Bill McDermott, described this as a transformative moment for businesses, revolutionising how work gets done. He expressed confidence that the AI Lighthouse customer program will inspire breakthrough ideas with significant “return on intelligence” or ROI.

Jensen Huang, the founder and CEO of NVIDIA, emphasised the rapid adoption of generative AI tools across industries and how this collaboration will help customers lead their respective fields by deploying generative AI solutions that leverage their own knowledge to transform everyday applications. NVIDIA has been actively investing and partnering with several companies across the globe for the rapid adoption of generative AI.

Since May, ServiceNow has introduced a range of generative AI capabilities designed for the Now Platform, tested in enterprise environments with leading pharmaceutical, manufacturing, and healthcare companies. ServiceNow’s Now Assist enables intelligent automation, increases productivity, and enhances user experiences by streamlining repetitive tasks and increasing agility using generative AI.

The AI Lighthouse Customer Program aims to build on this progress by collaborating with a select group of customers across IT service management (ITSM), customer service management (CSM), and employee experience to design, develop, and implement new generative AI use cases.

The post NVIDIA, Accenture, ServiceNow Launch AI Lighthouse for Enterprise appeared first on Analytics India Magazine.

Image created by Author with BlueWillow

Welcome to the inaugural edition of "This Week in AI" on KDnuggets. This curated weekly post aims to keep you abreast of the most compelling developments in the rapidly advancing world of artificial intelligence. From groundbreaking headlines that shape our understanding of AI's role in society to thought-provoking articles, insightful learning resources, and spotlighted research pushing the boundaries of our knowledge, this post provides a comprehensive overview of AI's current landscape. Without delving into the specifics just yet, expect to explore a plethora of diverse topics that reflect the vast and dynamic nature of AI. Remember, this is just the first of many weekly updates to come, designed to keep you updated and informed in this ever-evolving field. Stay tuned and happy reading!

Headlines

The "Headlines" section discusses the top news and developments from the past week in the field of artificial intelligence. The information ranges from governmental AI policies to technological advancements and corporate innovations in AI.

💡 AI Titans Pledge Responsible Innovation Under Biden-Harris Administration

The Biden-Harris Administration has secured voluntary commitments from seven leading AI companies — Amazon, Anthropic, Google, Inflection, Meta, Microsoft, and OpenAI — to ensure the safe, secure, and transparent development of AI technology. These commitments underscore three principles fundamental to the future of AI: safety, security, and trust. The companies have agreed to conduct internal and external security testing of their AI systems before release, share information on managing AI risks, and invest in cybersecurity. They also commit to developing technical mechanisms to ensure users know when content is AI-generated and to publicly report their AI systems' capabilities, limitations, and areas of appropriate and inappropriate use. This move is part of a broader commitment by the Biden-Harris Administration to ensure AI is developed safely and responsibly, and to protect Americans from harm and discrimination.

💡 Stability AI Unveils Stable Beluga: The New Workhorses of Open Access Language Models

Stability AI and its CarperAI lab have announced the launch of Stable Beluga 1 and Stable Beluga 2, two powerful, open access, Large Language Models (LLMs). These models, which demonstrate exceptional reasoning ability across varied benchmarks, are based on the original LLaMA 65B and LLaMA 2 70B foundation models respectively. Both models were fine-tuned with a new synthetically-generated dataset using Supervised Fine-Tune (SFT) in standard Alpaca format. The training for the Stable Beluga models was inspired by the methodology used by Microsoft in its paper: "Orca: Progressive Learning from Complex Explanation Traces of GPT-4.” Despite training on one-tenth the sample size of the original Orca paper, the Stable Beluga models demonstrate exceptional performance across various benchmarks. As of July 27th, 2023, Stable Beluga 2 is the top model on the leaderboard, and Stable Beluga 1 is fourth.

💡 Spotify CEO Hints at Future AI-Driven Personalization and Ad Capabilities

During Spotify's second-quarter earnings call, CEO Daniel Ek hinted at the potential introduction of additional AI-powered functionality to the streaming service. Ek suggested that AI could be used to create more personalized experiences, summarize podcasts, and generate ads. He highlighted the success of the recently launched DJ feature, which delivers a curated selection of music alongside AI-powered commentary about the tracks and artists. Ek also mentioned the potential use of generative AI to summarize podcasts, making it easier for users to discover new content. Furthermore, Ek discussed the possibility of AI-generated audio ads, which could significantly reduce the cost for advertisers to develop new ad formats. These comments come as Spotify seeks a patent for an AI-powered "text-to-speech synthesis" system, which can convert text into human-like speech audio that incorporates emotion and intention.

Articles

The "Articles" section presents an array of thought-provoking pieces on artificial intelligence. Each article dives deep into a specific topic, offering readers insights into various aspects of AI, including new techniques, revolutionary approaches, and ground-breaking tools.

📰 ChatGPT Code Interpreter: Do Data Science in Minutes

This KDnuggets article introduces the Code Interpreter plugin by ChatGPT, a tool that can analyze data, write Python code, and build machine-learning models. The author, Natassha Selvaraj, demonstrates how the plugin can be used to automate various data science workflows, including data summarization, exploratory data analysis, data preprocessing, and building machine-learning models. The Code Interpreter can also be used to explain, debug, and optimize code. Natassha emphasizes that while the tool is powerful and efficient, it should be used as a baseline for data science tasks, as it lacks domain-specific knowledge and cannot handle large datasets residing in SQL databases. Natassha suggests that entry-level data scientists and those aspiring to become one should learn how to leverage tools like Code Interpreter to make their work more efficient.

📰 Textbooks Are All You Need: A Revolutionary Approach to AI Training

This KDnuggets article discusses a new approach to AI training proposed by Microsoft researchers, which involves using a synthetic textbook instead of massive datasets. The researchers trained a model called Phi-1 entirely on a custom-made textbook and found that it performed impressively well in Python coding tasks, despite being significantly smaller than models like GPT-3. This suggests that the quality of training data can be as important as the size of the model. The Phi-1 model's performance also improved when fine-tuned with synthetic exercises and solutions, indicating that targeted fine-tuning can enhance a model's capabilities beyond the tasks it was specifically trained for. This suggests that this textbook-based approach could revolutionize AI training by shifting the focus from creating larger models to curating better training data.

📰 Latest Prompt Engineering Technique Inventively Transforms Imperfect Prompts Into Superb Interactions For Using Generative AI

The article discusses a new technique in prompt engineering that encourages the use of imperfect prompts. The author argues that the pursuit of perfect prompts can be counterproductive and that it's often more practical to aim for "good enough" prompts. Generative AI applications use probabilistic and statistical methods to parse prompts and generate responses. Therefore, even if the same prompt is used multiple times, the AI is likely to produce different responses each time. The author suggests that rather than striving for a perfect prompt, users should make use of imperfect prompts and aggregate them to create effective prompts. The article references a research study titled "Ask Me Anything: A Simple Strategy For Prompting Language Models" which proposes a method of turning imperfect prompts into robust ones by aggregating the predictions of multiple effective, yet imperfect, prompts.

Learning Resources

The "Learning Resources" section lists useful educational content for those eager to expand their knowledge in AI. The resources, ranging from comprehensive guides to specialized courses, cater to both beginners and seasoned professionals in the field of AI.

📚 LLM University by Cohere: Your Gateway to the World of Large Language Models

Cohere's LLM University is a comprehensive learning resource for developers interested in Natural Language Processing (NLP) and Large Language Models (LLMs). The curriculum is designed to provide a solid foundation in NLP and LLMs, and then build on this knowledge to develop practical applications. The curriculum is divided into four main modules: "What are Large Language Models?", "Text Representation with Cohere Endpoints", "Text Generation with Cohere Endpoints", and "Deployment". Whether you're a new machine learning engineer or an experienced developer looking to expand your skills, the LLM University by Cohere offers a comprehensive guide to the world of NLP and LLMs.

📚 Free From Google: Generative AI Learning Path

Google Cloud has released the Generative AI Learning Path, a collection of free courses that cover everything from the basics of Generative AI to more advanced tools like the Generative AI Studio. The learning path includes seven courses: "Introduction to Generative AI", "Introduction to Large Language Models", "Introduction to Image Generation", "Attention Mechanism", "Transformer Models and BERT Model", "Create Image Captioning Models", and "Introduction to Generative AI Studio". The courses cover a range of topics, including Large Language Models, Image Generation, Attention Mechanism, Transformer Models, BERT Model, and Image Captioning Models.

Research Spotlight

The "Research Spotlight" section highlights significant research in the realm of AI. The section includes breakthrough studies, exploring new theories, and discussing potential implications and future directions in the field of AI.

🔍 The Role of Large Language Models in the Evolution of Data Science Education

The research paper titled "The Role of Large Language Models in the Evolution of Data Science Education" discusses the transformative impact of Large Language Models (LLMs) on the roles and responsibilities of data scientists. The authors argue that the rise of LLMs is shifting the focus of data scientists from hands-on coding to managing and assessing analyses performed by automated AI systems. This shift necessitates a significant evolution in data science education, with a greater emphasis on cultivating diverse skillsets among students. These include creativity informed by LLMs, critical thinking, programming guided by AI, and interdisciplinary knowledge.

The authors also propose that LLMs can play a significant role in the classroom as interactive teaching and learning tools. They can contribute to personalized education and enrich learning experiences. However, the integration of LLMs into education requires careful consideration to balance the benefits of LLMs while fostering complementary human expertise and innovation. The paper suggests that the future of data science education will likely involve a symbiotic relationship between human learners and AI models, where both entities learn from and enhance each other's capabilities.

More On This Topic

• Towards a Responsible and Ethical AI
• Navigate the road to Responsible AI
• The Case for a Global Responsible AI Framework
• Coding Ethics for AI & AIOps: Designing Responsible AI Systems
• Put Responsible AI into Practice—attend the digital event on December…
• 8 Ways to Improve Your Search Application this Week

Social media has been slowly implementing generative AI into its models. However, it has primarily been focused on helping marketers and businesses create posts, as seen with LinkedIn and Meta.

Leaked screenshots shared by app researcher Alessandro Paluzzi on Twitter suggest that Meta is taking a different approach with Instagram and using AI to develop several features that would directly impact user experience on the app.

Instagram is working on labels that help users distinguish between AI-generated and real photos, a feature that could have a significant impact on the user experience, as well as help curb misinformation.

Also: DeepMind's RT-2 makes robot control a matter of AI chat

The feature would help ease users' concerns about the negative impacts of generative AI and discerning what is real or generated. (It could also help avoid another Pope Francis wearing a puffer jacket moment.)

Another feature Paluzzi shared on Twitter was direct message summaries, which could be especially useful for influencers and content creators who get bombarded with brand messages.

Instagram is also harnessing generative AI to make photo editing and getting the perfect picture easier.

One tool, called Restyle, would let users transform their images into any visual style they prompt. Another tool, AI brush, could be used to "add or replace specific parts of your image," per the screenshot shared.

The AI Brush tools would be similar to Samsung's Object Eraser feature or Google Pixel's Magic Eraser, which make it simple to remove unwanted objects from photos.

Also: How researchers broke ChatGPT and what it could mean for future AI development

With a growing influencer economy and dependency on social media, getting the perfect picture is more important than ever, and these tools help users achieve that in post-editing.

As previously covered by ZDNET, there have been reports of Instagram implementing its own AI chatbot, similar to Snapchat's My AI.

When these features will be available to the public remains unknown; however, Meta has been quickly adding to its generative AI tool arsenal, and we can expect these to be developed and deployed fairly quickly too.

Artificial Intelligence

Attend any keynote speech by a tech CEO and you can be sure of one thing: generative AI will get a mention.

Since the launch of Open AI's ChatGPT late last year, the IT industry has been fixated on the application and exploitation of artificial intelligence.

I went to an IT conference in London recently where the expo floor was dominated by a sales booth with signage at the top that said 'ChatGPT' in massive, flashing letters.

Also: Generative AI will soon go mainstream, say 9 out of 10 IT leaders

I think it was an attempt by the organizers to piggyback on the cacophonous hype of the big tech trend of the moment.

In fact, having attended the keynote speech at the start of the day — where the CEO made regular references to the importance of AI, but without making any product announcements — an attempt to put an AI-enabled stake in the ground was clearly the flavor of the event.

But this unnamed IT vendor is far from alone in its desire to be associated with all things generative AI.

Take Sundar Pichai and his senior colleagues at Google, who mentioned AI roughly 143 times over the course of the company's recent two-hour keynote presentation at Google I/O, according to CNET estimates.

If talking as much as possible about generative AI is good enough for Google, then you can bet your bottom dollar it's going to be good enough for every other IT executive.

However, if you're a professional who's charged with making sense of emerging technology, then you're going to need a way to sort the generative AI wheat from the chaff.

Also: Generative AI should be more inclusive as it evolves, according to OpenAI's CEO

And with every tech company trying to bolt on AI services, how can professionals work out which vendors and products will provide value to their businesses, both today and long into the future?

Four business leaders give us their best-practice tips.

1. Start your explorations now

Working out which vendor will provide the most value through AI is far from straightforward, says Tulia Plumettaz, director of machine learning at e-commerce giant Wayfair.

"There's a lot of speculation," says Plumetta. " would say at this point that it's extremely difficult to predict the direction that the field is going to take."

Despite all the uncertainty, Plumettaz recognizes professionals can't afford to sit back while rivals use generative AI to develop a competitive advantage.

Also: Most workers want to use generative AI to advance their careers but don't know how

For this reason, professionals should start exploring generative AI and look for vendor partners to investigate potential use cases.

Plumettaz explained to ZDNET how her company is working with Snorkel AI to boost the quality of the online search experience it provides to consumers — and just as Wayfair is dabbling in machine learning, so the company will also explore generative AI.

"I think there is going to be a lot of learning and bets on how AI will impact productivity," she says.

"There's also a lot of hypotheses — we know it's going to move fast. What are going to be the differentiators for the enterprise? Is it going to be about a few models ruling the field? That's one of the journeys that we are embarking on with Snorkel AI right now."

2. Focus on aims and cultures

Kavin Mistry, head of digital marketing and personalization at TSB Bank, is another professional who's helping his organization explore generative AI — and he also recognizes the cacophonous hype surrounding the subject.

"I share your horror in the sudden influx of AI-related pitches in the emails that I get sent."

Mistry's team is currently thinking about the use of AI as part of a much wider data strategy. As part of these efforts, the bank is working with Adobe.

Also: Today's AI boom will amplify social problems if we don't act now, says AI ethicist

He advises other professionals who are looking to sort the AI wheat from the chaff to focus on two core elements: aims and cultures.

"It's like any process. First, it's about understanding our need — our end state — and what we want our model to be," says Mistry.

"Then, we typically go through our procurement process to find the appropriate vendor. For me, it's important that you have the right-scale organization and technology that's aligned to your company's needs. There needs to be a strong cultural fit."

3. Stick to the classic rules

The focus on business requirements as the key to exploiting generative AI resonates with Wulstan Reeve, head of data marketplace at Legal & General Investment Management.

His team is using the Cloudera Data Platform to bring data together as a platform for insight-led experimentation in other areas, including potentially AI.

Reeve says any attempts to move into generative AI via vendor partnerships in the future will involve sticking to what he calls "classic rules".

Also: The future of cloud computing, from hybrid to edge to AI-powered

"It's all about business value. If there's no killer use case that you can start to apply, then playing is fine, right? But you're only ever going to get a certain amount of money to play," he says.

"The key is starting to make very tangible links to how you can generate revenue or reduce costs. So, I think a lot of these classic rules will continue to apply to this stuff. But it is right that lots of companies should be safely experimenting in AI."

4. Build some solid foundations

Lalo Luna, global head of strategy and insights at Heineken, says the brewing giant is using Stravito's technology to share insights through an internally branded platform known as Knowledge & Insight Management (KIM).

One of his team's priorities during the next few months is to think about how they start exploiting AI. Luna anticipates Stravito being an important partner in this journey.

"We are really looking to develop this capability. But we're a decentralized business and we have many different things that are happening all around the world," he says.

Also: The best AI chatbots: ChatGPT and other noteworthy alternatives

"So, what happens if you have information in many different places? If you want to take full advantage of AI, I think you need to connect all these places."

Stravito recently added a proprietary generative AI engine to try and improve the search experience for employees.

Luna says these kinds of enhancements should help his company to explore emerging technology such as generative AI from a solid, standardized base.

Also: The best AI art generators: DALL-E 2 and other fun alternatives to try

"When it comes to the democratization of insights for knowledge management, it's better to have one main platform," he says.

"We can connect everything and Stravito is going to be the heart of this ecosystem. We can then access different tools from the center, and we don't need to log into every specific tool to consult information."

Artificial Intelligence