Рубрика: AI

If I tell you that I saw a robot today, what comes to mind? What is a robot?

This is not a trick question.

Robots in science fiction

Because we've seen so many robots. We've seen Robbie the Robot from the 1956 movie Forbidden Planet, Rosie the housekeeper from The Jetsons, the animated Gigantor, C-3PO and R2-D2 from late 1970s Star Wars, Optimus Prime, Data from Star Trek, Arnold Schwarzenegger's T-800 Terminator, or later robots like Wall-E, Dolores (and all the synths) from Westworld, and all the rest of the robots in the various Star Wars spin-offs.

All of these, together, have built up our view of robots over the years, at least in movies and TV.

We're also familiar with the stories these robots tell. Data from Star Trek just wants to be more human. Isaac from The Orville is a Kaylon, a race of robots that destroy organic creatures. (Yet Isaac's path has been one of redemption, for he's a compassionate Kaylon and helps turn the tide for organics.) The Star Wars robots, especially those designed for merchandising, have become friends and companions to their organic buddies.

Then there are the many evil robots bent on destruction, like Ultron; Hal 9000; the Daleks and the Cybermen from Doctor Who; various incarnations of Terminator robots; Nomad, Lore, Peanut Hamper, and Control from Star Trek; and a collection of droids from Star Wars.

These robots have all provided writers with the opportunity to reflect humanity's traits and problems back on mechanical beings and to play with what happens when you create artificial life with or without the moral constraints that govern most humans.

Also: The Star Wars starter guide: Every movie ranked and graded

Robots in the real world

But robots exist in the real world. And they don't behave like C-3PO or Mr. Data. Instead, they range from giant automated factories to automobile welding robots, from 3D printers to toys for kids. What makes these robots…robots? And what makes them different from the robots of science fiction?

Foe starters, the robots of science fiction are often fully autonomous. Mr. Data from Star Trek: The Next Generation and The Doctor from Star Trek Voyager (a holographic AI) were even declared to be legal people in the eyes of the fictional Federation. Nobody is claiming that my friend's Tesla is legally a person.

Also: The best robots and AI innovations at CES

In fact, the auto industry has developed a set of aspirational criteria defining the "autonomous-ness" of a robotic vehicle, and we can apply that criteria to other robots as well. The SAE J3016 criteria have six levels: levels 0, 1, and 2 describe automation limits, while levels 3, 4, and 5 describe more fully autonomous functioning:

As you can see, the blue criteria specify that a human driver must still be in control, even if assisted by the car, while the green criteria specify (mostly) that the car is able to make all necessary decisions.

Most real-world robots we have today fall on the blue side of the spectrum. That's why I use the terms automated vs. autonomous to differentiate robotic capabilities. Although the words sound similar, here's how they differ:

• Automated systems follow pre-defined rules to perform specific tasks.
• Autonomous systems can operate independently, make decisions, and adapt to new situations.

Today, I contend, most robots are merely automated devices that have some level of movement in the real world. They perform a series of steps, possibly modified based on certain criteria. (For example, a 3D printer will stop printing when it runs out of filament, only to resume once more material is loaded.) Autonomous devices include C-3PO, Mr. Data, the T-800, or Amazon's dream for people-free delivery robots.

Also: BMW tests next-gen LiDAR to beat Tesla to Level 3 self-driving cars

Right now, we do automated really, really well. Autonomous, not so much. But we're getting there.

Enormous dynamic range

Today's robots, despite not being as versatile as Mr. Data, are generally quite useful and functional.

These include industrial robots, medical robots, military and defense robots, domestic robots, entertainment robots, space exploration and maintenance robots, agricultural robots, retail robots, underwater robots, and telepresence robots that help people participate in an activity from a distance.

My personal interest has been focused on robots available and accessible to makers and hobbyists, robots that can empower individuals to build, design, and prototype projects previously only feasible by those with a shop full of fabrication machinery.

I'm talking about 3D printers, which build up objects from layers of molten plastic; CNC devices, which often cut, carve, and remove wood or metal to create objects; laser cutters, which are ideal for sign cutting, engraving, and fabricating very detailed parts and circuit boards; and even vinyl cutters, for carefully cutting light, flexible material in intricate patterns.

Also: This is the best and fastest sub-$300 3D printer I've tested yet

These machines are programmed using CAD software to define — aka, design — the object being built. Those designs are then converted to a series of motion instructions that guide the machine in making repetitive, complex moves.

I used a CNC, for example, to make a series of identical custom organizer racks for parts storage.

While I designed and assembled the organizer, the robot became a force multiplier, carving precise rack-holding features, a process that was well beyond my woodworking skill set, which is mostly limited to hammering nails and screwing screws.

Robots today have an enormous dynamic range, from children's learning toys to something as astonishingly complex and enormous as Amazon's smart warehouses. These warehouses each contain thousands of robots, but because they all work in concert with each other, the entire warehouse can, itself, be considered a giant robot all on its own.

Span of autonomy

Let's return to our discussion of automated robots vs. autonomous robots. Automated robots can follow a set of tasks, usually overseen (or at least checked on regularly) by a human operator.

My 3D printers are a good example. When I create a design in CAD software and then convert that design into gcode, what I'm creating is a series of movement instructions. Instructions specify the X and Y position of the print head, along with how high the extruder needs to be to account for the layers that are being constantly added. Instructions also specify the temperature of the extruder, determining how quickly and smoothly the plastic melts onto the previous layer.

Usually, I'll kick off a print and then monitor it via a camera. On the not-rare-enough occasion that the print fails or the printer just decides to spew molten plastic into the air, I usually catch it fast enough, rush into the Fab Lab, and cancel the print. The printer is automated — it's following instructions — but there's nothing autonomous about this process.

Also: Generative AI will far surpass what ChatGPT can do

Newer printers are incorporating some AI: The cameras feed images to a processor that uses some machine learning to examine each image and determine if there's a failure situation. While the machines can't fix those failures, machine learning can turn off the process, preventing loss of material and a possible safety hazard.

When you visit Amazon's factories, you'll see more robots. There are carting robots that move along the floor, delivering products. Most of these are automated, not autonomous. But Amazon's wildly complex conveyor systems do have intelligent imaging systems that look at products as they pass by and make some decisions about the objects as they pass. Here, we're starting to see signs of more management taking place without human supervision.

I have a drone that also exhibits some autonomous behaviors. If, while directing it via a hand-held controller, I send it out of radio range, the drone itself will take over. It will plot a course back to its origin, reverse course, avoid obstacles like trees and power lines, and bring itself back home without any interaction. It will perform the same behavior when it senses its battery is too low for it to continue flying.

In each of these three examples (AI-assisted 3D printer monitoring, warehouse conveyor monitoring, and return-to-home flight), we're seeing autonomous behaviors built as an extension of a mostly automated system. I think that's how we'll see autonomous features roll out. They'll be available for situation-by-situation until more and more situations are taken into account.

Also: The tech behind ChatGPT could power your next car's AI driving assistant

Eventually, you'll be able to crawl into your car and get an extra 45 minutes of shuteye while the vehicle drives you to the Starbucks nearest your office. But as the SAE Levels of Driving Automation chart we discussed earlier shows, Level 5 is a big step. At that point, we're trusting the vehicle to handle any and all road conditions and respond intelligently, carefully, quickly, and safely. Most experts believe that we'll start seeing cars with this capability sometime after 2030.

When ChatGPT blunders with one of its famous hallucinations, it's merely annoying — and possibly embarrassing if someone uses that material in some writing. But if a robot blunders while operating in the real world, something can go wrong physically, even fatally. Because the stakes are high, much more care needs to be taken not only in the development of fully autonomous systems, but in the staging and release of those systems in order to make sure they're safe to unleash on in our shared environment.

Looking forward: Robots of tomorrow

Let's review our three main takeaways: First, science fiction has given us a picture of a robot that is both cautionary and aspirational — but not necessarily practical. Second, a great many things can be considered robots in the real world. And third, the range of autonomy can vary among different real-world robots.

At first glance, it seems as though AI and robotics are inextricably linked. But as we've seen, AI can inform all, part, or none of a robot's function, depending on the level of technology involved and the purpose of a robot. While it would be nice for a fabrication robot to know when it is failing and stopping, we derive a great deal of value from automated CNC devices and 3D printers that just follow their gcode instructions.

As we look into the future, we'll see more autonomous systems. Siemens has a fascinating vision of what a factory will look like in the coming decades, and it showcases many autonomous systems interacting with the production process overall.

Also: How horses can inform the future of robot-human interaction

Outside of the world of entertainment characters, robots are complex mechanisms that justify the cost and effort to create them by the value they generate, whether that be cost savings, time savings, the ability to take on otherwise difficult processes, the ability to operate in environments dangerous to humans, or the ability to force-multiply the efforts of their human operators.

The ability to interact with the real world and perform automated steps are table stakes for participating in the robotics revolution. As we move forward, expect a melding between machine learning, intelligent vision, generative AI, traditional programming skills, and mechanical design prowess to open up new doors, provide new opportunities, and help robots of all sizes and capabilities do more to help us.

On the other hand, if — someday in the future — robots start to yell, "Exterminate! Exterminate!" …well, then… Danger, Will Robinson, Biddi Biddi Biddi, These are not the droids you're looking for.

Hasta la vista, baby!

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.

Nvidia Showcases Domain-specific LLM for Chip Design at ICCAD November 2, 2023 by John Russell

Nvidia H100 die.

This week Nvidia released a paper demonstrating how generative AI can be used in semiconductor design. Nvidia chief scientist Bill Dally announced the new paper during his keynote at the International Conference on Computer-Aided Design (ICCAD) now taking place in San Francisco.

“This effort marks an important first step in applying LLMs to the complex work of designing semiconductors,” said Dally at the event in San Francisco. “It shows how even highly specialized fields can use their internal data to train useful generative AI models.”

Mark Ren, an Nvidia research director and lead author on the paper, said “I believe over time large language models will help all the processes, across the board. Nvidia issued a blog on the work along with the paper (ChipNeMo: Domain-Adapted LLMs for Chip Design).

Designing today’s giant chips, such as Nvidia’s H100 GPU, is typically a two-year effort involving multiple engineering teams. The sudden emergence of LLM and Generative AI has triggered a wave of efforts to develop customized, domain-specific LLMs. Bloomberg’s financial LLM (BloombergGPT) is a good example.

The expectation is that domain-specific LLMS will join the EDA tool world and significantly speed and improve complex chip design. At this point ChipNeMo is an internal project for internal use only. The paper’s abstract summarizes the work nicely:

Abstract: "ChipNeMo aims to explore the applications of large language models (LLMs) for industrial chip design. Instead of directly deploying off-the-shelf commercial or open-source LLMs, we instead adopt the following domain adaptation techniques: custom tokenizers, domain-adaptive continued pretraining, supervised fine-tuning (SFT) with domain-specific instructions, and domain-adapted retrieval models.

"We evaluate these methods on three selected LLM applications for chip design: an engineering assistant chatbot, EDA script generation, and bug summarization and analysis. Our results show that these domain adaptation techniques enable significant LLM performance improvements over general-purpose base models across the three evaluated applications, enabling up to 5x model size reduction with similar or better performance on a range of design tasks. Our findings also indicate that there’s still room for improvement between our current results and ideal outcomes. We believe that further investigation of domain-adapted LLM approaches will help close this gap in the future."

The paper authors write, “We believe that LLMs have the potential to help chip design productivity by using generative AI to automate many language- related chip design tasks such as code generation, responses to engineering questions via a natural language interface, analysis and report generation, and bug triage.”

No doubt Nvidia has self-interest here, both in strengthening its position in LLM development/provider community as well as stirring demand for its broad product portfolio. ChipNeMo was built using Nvidia’s NeMo cloud framework for LLM development and training.

Before starting the ChipNeMo project, Nvidia conducted a survey of potential LLM applications within own design teams. According to the paper, the responses fell roughly into four buckets: code generation, question & answer, analysis and reporting, and triage.

“Code generation refers to LLM generating design code, testbenches, assertions, internal tools scripts, etc.; Q & A refers to an LLM answering questions about designs, tools, infrastructures, etc.; Analysis and reporting refers to an LLM analyzing data and providing reports; triage refers to an LLM helping debug design or tool problems given logs and reports. We selected one key application from each category to study in this work, except for the triage category which we leave for further research,” according to the paper.

The paper walks through the strategy and steps taken in developing ChipNeMo, providing a rough template for others. Obviously, there are more tasks that could be tackled. In the blog Nvidia reported it has other semiconductor design projects using AI to design smaller, faster circuits and to optimize placement of large blocks.

One important lesson learned from ChipNeMo project, reported Nvidia, is that these domain-specific LLMs can be substantially small and run effectively on smaller compute platforms.

This from the bog: “On chip-design tasks, custom ChipNeMo models with as few as 13 billion parameters match or exceed performance of even much larger general-purpose LLMs like LLaMA2 with 70 billion parameters. In some use cases, ChipNeMo models were dramatically better. Along the way, users need to exercise care in what data they collect and how they clean it for use in training, Ren added.”

Link to blog, https://blogs.nvidia.com/blog/2023/10/30/llm-semiconductors-chip-nemo/

Link to paper, https://d1qx31qr3h6wln.cloudfront.net/publications/ChipNeMo%20%2824%29.pdf

Related

Stability AI’s latest tool uses AI to generate 3D models Kyle Wiggers 9 hours

Stability AI, the startup behind the text-to-image AI model Stable Diffusion, thinks 3D model creation tools could be the next big thing in generative AI.

At least, that’s the message it’s sending with the launch of Stable 3D, an AI-powered app that generates textured 3D objects for modeling and game development platforms like Blender, Maya, Unreal Engine and Unity.

Available in private preview for select customers who reach out to Stability via the company’s contact form, Stable 3D is designed to enable non-experts to generate “draft-quality” 3D models “in minutes,” Stability AI writes on its blog.

“For graphic designers, digital artists and game developers, 3D content creation can be among the most complex and time-consuming tasks, often taking hours — sometimes days — to create a moderately complex 3D object,” writes the company. “Stable 3D levels the playing field for independent designers, artists and developers, enabling them to create thousands of 3D objects per day at very little cost.”

Bombast aside, Stable 3D seems fairly robust — and comparable in terms of its capabilities to other model-generating tools on the market. Users can describe in natural language a 3D model they want to create or upload an existing image or illustration to convert into a model. Stable 3D outputs 3D models in the “.obj” file format, which allows them to be edited and manipulated using most standard 3D modeling tools.

Stability hasn’t revealed which data it used to train Stable 3D. Given generative AI models’ tendency to regurgitate training data, this could become a point of concern down the line for the tool’s commercial users. If any of the data was copyrighted and Stability AI didn’t obtain the proper licensing, Stable 3D customers could end up unwittingly incorporating IP-infringing work into their projects.

And Stability AI doesn’t have the best track record when it comes to respecting IP. Earlier this year, Getty and several artists sued the startup for allegedly copying and processing millions of images owned by them to train Stable Diffusion without proper notification or compensation.

Stability AI recently partnered with startup Spawning to respect “opt-out” requests from artists, but it’s unclear if the partnership covers Stable 3D’s training data. We’ve reached out to Stability AI for more information and will update this post if we hear back.

Potential legal ramifications aside, Stable 3D marks Stability AI’s entrance into the nascent — but already crowded — field of AI-powered 3D model generation.

3D models generated with Stability AI’s new Stable 3D tool.

There’s 3D object-creating platforms such as 3DFY and Scenario, as well as startups like Kaedim, Auctoria, Mirage, Luma and Hypothetic. Even incumbents such as Autodesk and Nvidia are beginning to dip their toes in the space with apps like Get3D, which converts images to 3D models, and ClipForge, which generates models from text descriptions.

Meta, too, has experimented with tech to generate 3D assets from prompts. So has OpenAI, which last December released Point-E, an AI that synthesizes 3D models with potential applications in 3D printing, game design and animation.

Stable 3D appears to be Stability AI’s latest attempt to diversify its business — or perhaps pivot — in the face of increasing competition from art-creating generative AI platforms including Midjourney and the aforementioned OpenAI.

In April, Semafor reported that Stability AI was burning through cash — spurring an executive hunt to ramp up sales. According to Forbes, the company has repeatedly delayed or outright not paid wages and payroll taxes, leading AWS — which Stability uses for compute to train its models — to threaten to revoke Stability’s access to its GPU instances.

Stability AI recently raised $25 million through a convertible note (i.e. debt that converts to equity), bringing its warchest to more than $125 million. But it hasn’t closed new funding at a higher valuation; the startup was last valued at $1 billion. Stability was said to be seeking quadruple that figure within the next few months despite stubbornly low revenues.

In other seeming attempts at differentiation and drumming up sales, Stability AI today announced new features for its online AI-powered photo editing suite, including a model fine-tuning feature that lets users personalize the underlying art-generating models and a “sky replacer” tool that substitutes the color and aesthetic of the sky in photos with preset alternatives.

The new tools join Stability AI’s growing stable of AI-powered products, including the music-generating suite Stable Audio, doodle-creating app Stable Doodle and a ChatGPT-like chatbot.

Everyday there is a new acronym popping up in the generative AI world. One of the latest buzzwords is RAG, which stands for retrieval augmented generation. And it is not just another acronym; it represents a significant leap in the field of LLMs. But what exactly is it?

RAG has gained popularity because it combines the strengths of both retrieval-based and generative-based models. Basically, RAG is about attaching a new database onto a base model, and letting the model retrieve new information from it, and then generate information from it.

This helps in reducing the hallucination in the model. Mostly, it is a vector database, or in certain cases such as GPT-4, it is the internet.

At Cypher 2023, Dhruv Motwani, founder & CEO at SpringtownAI, also discussed and demonstrated the use of RAG and explored its architecture and functionality. The participants were able to deploy applications on their AWS accounts and test the hallucinations of the models, which were lesser when using RAG.

Is it really any good?

“The RAG that you see today is really glorified prompt engineering. The most common standard RAG flow we currently have is completely unaware of the context of your data. Without looking up in your data, it sends the lookup plus the original query to GPT,” Mark McQuade, co-founder of Arcee.ai told AIM.

McQuade and his team have built an End2End RAG system called DALM, which sits on top of the main LLM. He said that the best way to use it is to pair it with an in-domain specialised model with the system, instead of a larger model with tons of unnecessary data.

Before delving into RAG, it’s essential to understand the choices that AI developers have when working with AI models. They can either build a model from scratch, fine-tune an existing model, or employ retrieval augmented generation. Each approach has its pros and cons, and the bigger the model, the bigger are the chances of hallucinations.

Building from the ground up can be a costly and time-consuming endeavour. For instance, OpenAI invested over $100 million to train its GPT-4 model. On the other hand, fine-tuning existing models with additional data is a viable option, but it carries the risk of the model “forgetting” some of its original training data.

RAG combines retrieval-based and generative-based AI models to deliver context-aware responses. A retrieval model is used to access information from existing knowledge sources, such as databases or online articles. The generative model then takes this retrieved information and synthesises it into coherent, contextually appropriate responses.

The key advantage of RAG is its ability to provide responses that are not only accurate but also unique, akin to human language, rather than simply summarising retrieved data.

At its core, RAG is essentially an advanced form of prompt engineering. It focuses on keeping the model fixed and optimising how it “stuffs” the context window with text to answer specific questions. This approach is particularly beneficial for prompt engineers who need to learn AI engineering skills and master a base set of prompts to build RAG/agent systems.

In more complex RAG/agent systems, it’s not just about a single prompt; it involves a collection of prompts that work in harmony to provide accurate and context-aware responses.

RAG extended

Some researchers argue that RAG might not be as beneficial as compared to a longer context window, as both of these offer the same results. A recent study titled ‘Retrieval meets Long Context Large Language Models‘ compared RAG with longer context window LLMs.

The paper found out that open source Embedding Models/Retrievers outperformed OpenAI models. Combining a simple RAG with a 4k LLM could match the performance of long context LLM. Moreover, RAG paired with a 32k LLM outperformed providing the full context.

While RAG offers significant benefits, it’s important to consider the potential for bad responses when retrieving information. Philipp Schmid, tech lead at Hugging Face questioned if we teach LLMs to be more factually correct and self-reliable and introduced Self-RAG. It is a novel approach of teaching models when to retrieve information and how to use it effectively.

Self-RAG involves creating a “critique” dataset to determine when retrieval is appropriate and what information is relevant. By creating a “critique” dataset with retrieval guidelines, developers can then train the critique model on the synthetic dataset. Using prompts, the critique model, and a retriever, the developer can generate a RAG dataset.

After training the LLM on the RAG dataset, including special tokens to instruct the model on when to retrieve or generate responses, the model during inference adaptively generates special tokens based on the query to determine whether retrieval is necessary.

It seems like the hallucinations will continue for a while. But just like prompt engineering, it is time to learn the next AI engineering skill, and catch up with the buzzword, and this time it is to build a base set of prompts for RAG systems.

The post RAG is Just Fancier Prompt Engineering appeared first on Analytics India Magazine.

Google Vertex AI Search Add News GenAI Capabilities And Enterprise-Ready Features November 2, 2023 by Ali Azhar

Generative AI (GenAI) has ushered in a new era of interactive and multimodal experience for developers, businesses, and governments. For GenAI to achieve its potential, it needs to be easily accessible and integrated into a range of applications so users with little machine learning expertise can develop and deploy intelligence apps.

To achieve this goal, Google had earlier unveiled its vision for the Vertex AI Search and made it generally available in August. The company has now announced new GenAI updates to Google Cloud’s Vertex AI Search. The enhancements will empower customers to get their applications off the ground faster, increase productivity, uncover hidden insights across data, and drive greater user satisfaction.

Google Vertex AI Search provides customers with a unified platform to manage end-to-end ML workflows, use and fine-tune pre-build models, and harness the power of Google Cloud’s infrastructure to scale resources.

The recently announced beta launch of Adelaide by Forbes is an example of the value offered by Google Vertex AI. Adelaide is a GenAI search platform that combines Forbes trusted journalism with AI-driven personalized recommendations. The search and conversation capabilities offered by Google Vertex AI help make content discovery easier and more intuitive for Forbes audiences.

“As we look to the future, we are enabling our audiences to better understand how AI can be a tool for good and enhance their lives," said Vadim Supitskiy, Chief Digital and Information Officer, Forbes. "Adelaide is poised to revolutionize how Forbes audiences engage with news and media content, offering a more personalized and insightful experience from start to finish.”

Customizable Answers and Search Tuning

The new GenAI capabilities include tailored search to fit business needs, especially for large enterprises that need highly customized AI-driven search. Developers now have more control over the prompts including the length, tone, style, and format of the information presented. In addition, they can let the users customize the output through different options presented through a drop-down menu. For example, users can choose to view a “standard” or “simple” prompt.

Along with enhanced capabilities for building apps, Vertex AI Search now allows users to use their own data for search to boost the accuracy of results. Even small training sets can allow Google Vertex AI Search to refine its rankings and deliver better search experiences. There is even an option for users to build their own GenAI applications for more complex use cases using Vertex AI Embeddings.

Vertex AI Embeddings

Vertex AI now offers a set of embedding models to support use cases for semantic search, classification, outlier detection, and recommendations. The embeddings can be uploaded to Vector Search to pair with other Vertex AI foundation models and services to power predictive and generative AI applications.

Vertex AI’s Text Embeddings and Multimodal Embeddings (supporting Text and image) models are both generally available, and new Multimodal Embeddings models that support text, image, and now video are being announced in preview.

Vertex Search

The Vertex Search, formerly known as Machine Machine Matching, can index data as vector embeddings to find the most relevant embeddings at scale. It uses a search algorithm called approximate nearest neighbor (ANN) that can handle high throughput while providing high recall at low latency. The UI has also been updated to minimize the need for any coding. In addition, the index time has been reduced and filter capabilities have been enhanced.

Grounding Data

A key challenge for enterprises that use GenAI is that foundation models can perceive objects or patterns that are imperceptible to human observers, and this creates inaccurate outputs. To help prevent this “AI hallucination”, Vertex AI Search offers multiple options to ground data.

With its new capabilities, Vertex AI Search offers enterprises a method for grounding in their own enterprise data to help users verify and validate results across disparate data sources. They can also use Vertex AI Connectors to expand data sources to other enterprise applications. There is also a new option for users to leverage wide sources of information for discovery needs to minimize the time and effort required for searching multiple sources for the same data.

Compliance-First Search

Google Vertex AI supports a range of compliance and security standards including ISO 27000-series, HIPAA, and SOC-1/2/3. These standards help ensure the integrity, transparency, confidentiality, and accountability of your data.

Google has now announced expanding support for access transparency to help enhance visibility and control over your cloud provider with admin access logs and approval controls. It will also provide customers with better awareness of Google administers access to their data. The customer-managed encryption keys (CMEK) are now available in preview. This new capability allows customers to encrypt their core content with their own encryption keys.

Related Items

Google Extends Vertex with More GenAI Features

AI in the Cloud: Google, Microsoft, and Amazon’s Divergent Strategies

Duet AI Goes Everywhere in Google’s Cloud

Related

Microsoft Research India is creating Shiksha Copilot, a generative AI tool to aid teachers in crafting customised learning experiences, designing assignments, and generating interactive activities. It combines various content types, including text, images, videos, charts, and interactive elements, while emphasising proficiency in multiple languages.

It is being developed in collaboration with the Sikshana Foundation, a local non-profit organisation dedicated to enhancing public education, and is currently being tested in more than 10 public schools in Bengaluru.

The team reports significant improvements, as teachers can now create comprehensive lesson plans in just 60-90 seconds instead of the previous 60-90 minutes. This initiative is part of Project VeLLM (Universal Empowerment with Large Language Models) at Microsoft Research India, which primarily aims to address the digital divide by overcoming language, income, digital literacy, and information access barriers with LLMs.

Project VeLLM strives to create an inclusive approach for LLM applications that can benefit people of diverse languages and cultures globally, making this technology accessible to a broader and more diverse population.

Designing Shiksha Copilot

Designing and building Shiksha Copilot involves handling diverse educational content types like text, images, videos, charts, and interactive elements. The goal is to create generative AI models with unified multimodal capabilities, particularly focusing on enhancing multilingual support. Shiksha Copilot offers several features to address these challenges, aligning content with specific curricula and learning objectives.

This is achieved by leveraging optical character recognition, computer vision, and generative AI models while supporting natural language and voice-based interactions for English and Kannada speakers. It connects to public and private educational resources and can be accessed through platforms like WhatsApp, Telegram, and web applications.

Semantic caching with LLMs is utilised to expedite content creation efficiently and reduce computational demands. The project maintains a strong focus on safety, reliability, and trustworthiness, implementing rigorous responsible AI procedures and content filtering to ensure the production of factual and reliable educational content.

In September, OpenAI announced a new guide for teachers to use ChatGPT in classrooms. The newly released guide contains recommended prompts, an overview of ChatGPT’s functioning and limitations, besides the efficacy of AI detectors, and a discussion on biases.

All this while students were using LLM based chatbots like OpenAI’s ChatGPT, Google’s Bard to complete homework assignments, learn about different topics and more. However, now the tables have turned and this new copilot is going to boost the productivity of teachers.

Read more: After Getting Banned in Schools, OpenAI Launches ChatGPT Tool for Teachers

The post Microsoft Unveils Shiksha Copilot for Teachers in India appeared first on Analytics India Magazine.