Рубрика: AI

GalaxEye Space, a space-tech startup, has formally signed an agreement with iDEX DIO to design and develop a multi-sensor fusion processing system for miniaturised satellites capable of carrying payloads up to 150 kg for the Indian Air Force.

The innovative system will analyse data from SAR, Earth Observation (EO), IR, and Hyper-Spectral sensors before transmitting it to ground stations. Designed as an independent, pluggable module, the device can be attached to any multi-sensor satellite, providing high processing throughput, power efficiency, compact size, lightweight reliability, compatibility with existing systems, and ease of maintenance.

The formal signing ceremony was attended by Giridhar Aramane, Hon’ble Defence Secretary, Lt Gen Upendra Dwivedi, designated Chief of Army Staff and current Vice Chief, and other high-ranking officials from the Ministry of Defence and the Armed Forces.

This collaboration aligns with the Make In India initiative, demonstrating GalaxEye Space’s commitment to fostering self-reliance in the defense sector, with support from iDEX DIO. The 350th iDEX contract enables innovation in space electronics, wherein many payloads earlier deployed on large satellites are now being miniaturised.

GalaxEye Space is also on track to launch the world’s first multi-sensor SAR + EO satellite by Q1 of FY 2025, marking a significant milestone in the industry. The company has developed India’s First UAV SAR system for defense markets and completed over 200 successful flights with UAV SAR Payload.

GalaxEye Space, a spin-out from team Hyperloop of IIT Madras, has been featured amongst the Top 10 Start-ups to watch in 2023 by ViaSat.

Bengaluru-based Pixxel had also recently signed the 350th contract under the iDEX (Innovations for Defence Excellence) program to manufacture miniaturised multi-payload satellites for the Indian Air Force.

The contract, awarded as part of the iDEX Prime Space grant, marks a significant milestone in Pixxel’s mission to revolutionise the space industry in India.

Under the contract, Pixxel will develop small satellites weighing up to 150 kg for electro-optical, infrared, synthetic aperture radar, and hyperspectral applications. The company will leverage its indigenous hyperspectral satellite technology and manufacturing expertise to build these satellites, enabling ease of manufacture, low cost, and ease of launch.

The post GalaxEye Inks iDEX Agreement to Develop Multi-Sensor Fusion System for IAF Satellites appeared first on Analytics India Magazine.

Writing down notes with pen and paper scratches an itch that typing text doesn't fulfill. Studies have shown that physically jotting down notes is a better way to retain information. However, by writing stuff down, you miss out on the perks of digitizing your notes. ChatGPT helps you get the best of both worlds.

When OpenAI supercharged the free version of ChatGPT with GPT-4o in May, users could upload files, including images, documents, and more. This update allowed users to interact with images in multiple ways, including extracting text.

Also: How to use ChatGPT to analyze PDFs for free

This capability means you can upload handwritten documents, from sticky notes to meeting and class notes to packing lists, and convert them into text. Then, you can use that text to create new content by copying and pasting it into presentations, emails, outlines, essays, Quizlets, and more.

Sound too good to be true? I thought the same, but after testing the tool multiple times, I can assure you that it works efficiently and quickly. Getting started is simple, and you will not want to stop once you start.

1. Log in to your OpenAI account

Even though you can access ChatGPT without creating an OpenAI account, you need to sign in to access GPT-4o and its many perks, including image uploads. The good news is that creating an account is easy, and the perks are well worth it.

Also: Grammarly adds 5 new security and control features for enterprise users

If you have never created a ChatGPT account, you can easily do so from the sign-in page. You can also log in with your existing Google or Microsoft account. I opted for the latter option, so I don't have to memorize another username and password.

2. Upload your image

Once you log in, you will be brought to the ChatGPT interface with a blank textbox. Next to the textbox, you will find a paperclip icon, which you can click on to upload your photo from several different sources, including Google Drive, Microsoft OneDrive, or your computer.

If the image is readily available on your device, you can drag and drop it to your text box.

Also: Gmail users can now ask Google's Gemini AI to help compose and summarize emails

I wouldn't worry about what your text looks like because I uploaded a sticky note I wrote over the weekend in cursive, as seen below, and ChatGPT had no problem understanding and outputting the text.

If you enter the picture with no text prompt, ChatGPT will most often automatically extract the text and present it to you typed out. However, to ensure the bot knows what to do and does so successfully, I recommend adding a simple text prompt like, "Can you extract the text from this image?" and then hit enter.

3. Use your text for whatever you need

Once ChatGPT outputs the extracted text, you can check to ensure the output accurately represents your text. In every instance I have tried, the bot has been word-for-word correct. Then, you can copy and paste the text wherever you'd like. Some use cases include presentations, emails, virtual sticky notes, outlines, and more.

Artificial Intelligence

Researchers from Tsinghua University and Shanghai AI Laboratory have unveiled MotionBooth, a groundbreaking generalist motion generation model capable of producing diverse and realistic human-object interactions.

This innovative approach addresses the limitations of existing motion generation methods, which often struggle with complex interactions and lack generalisation capabilities.

Read the full paper here – https://arxiv.org/pdf/2406.17758

MotionBooth employs a novel architecture that combines a motion transformer with a latent diffusion model, enabling it to generate high-quality motions for various human-object interaction scenarios.

The model’s key features include a unified representation for both human and object motions, a motion transformer that captures temporal dependencies, and a latent diffusion model for generating diverse and realistic motions.

The researchers trained MotionBooth on a large-scale dataset comprising over 200,000 motion sequences, including both human-only and human-object interaction motions. This extensive training allows the model to generalise well to unseen objects and interactions.

Experimental results demonstrate MotionBooth’s superior performance compared to existing methods, showcasing higher-quality motion generation for both seen and unseen objects, improved diversity in generated motions, and better generalisation to novel interaction scenarios.

The model’s capabilities extend to various applications, including motion synthesis for animation and game development, human-robot interaction design, and virtual reality and augmented reality experiences.

MotionBooth’s ability to generate realistic and diverse human-object interactions represents a significant advancement in the field of motion generation. This breakthrough has the potential to revolutionise industries relying on realistic motion synthesis, from entertainment to robotics.

As research in this area continues, future work may focus on further improving the model’s generalisation capabilities and expanding its applications to more complex scenarios involving multiple humans and objects.

This is the latest from Tsinghua University. A few days back, the university introduced the ChatGLM Model, which exceeds the capabilities of GPT-4 across a wide range of benchmarks and tasks.

The post MotionBooth AI Model Brings Realistic Motion to Virtual Worlds appeared first on Analytics India Magazine.

AWS has announced a collaboration with EvolutionaryScale to bring their cutting-edge language models to scientists and researchers in the field of biology. This partnership aims to advance applications in drug discovery, carbon capture, and more.

The collaboration introduces EvolutionaryScale’s ESM3, a state-of-the-art language model family, to AWS’s robust infrastructure. This includes enterprise-grade security, privacy measures, and purpose-built services for health and generative AI.

The ESM3 model family, which includes generative, multimodal models, will be available on AWS platforms such as Amazon SageMaker and AWS HealthOmics, with support for Amazon Bedrock coming later this year.

ESM3 enables researchers to generate complex multi-domain proteins from scratch, create protein design workflows, and incorporate functional understanding. This innovative approach, termed “programmable biology,” has the potential to significantly reduce the time and cost of bringing new therapeutics to market.

Customers can start using ESM3 through Amazon SageMaker and orchestrate automated drug discovery workflows via AWS HealthOmics. The collaboration also leverages AWS’s generative AI infrastructure, including high-performance GPU instances and ML accelerators like AWS Trainium and AWS Inferentia, ensuring efficient training and deployment of ESM3 models.

This initiative marks a significant leap in generative AI for biology, potentially accelerating drug discovery timelines and enabling novel therapeutic developments. The collaboration underscores AWS’s commitment to advancing generative AI across diverse industries, particularly in life sciences.

This is just the latest in AWS updates. Only a few days ago, AWS also launched a new AWS Generative AI Spotlight programme in the Asia Pacific and Japan (APJ) region, a four-week accelerator aimed at supporting early-stage startups in the region developing generative AI applications. In India, AWS is collaborating with venture capital firm Accel for this initiative.

The post AWS Brings ESM3 Language Models to Life Sciences appeared first on Analytics India Magazine.

Organizations are still choosing to over-provision their data center and IT requirements, which can get in the way of efforts to run these facilities more sustainably.

It is not uncommon, for instance, for enterprises to turn off operating features that allow a system to run at its most efficient mode, said John Frey, chief technologist of sustainable transformation at Hewlett Packard Enterprise (HPE).

Also: Apple is building a high-security OS to run its AI data centers — here's what we know so far

He explained that HPE ships its devices set to operate at their most efficient level, with power performance optimized. However, customers often will turn off the default setting as soon as they receive the new product, worried that it would deprive the system of the computing power it needs to run without lag.

"So we can design our products to operate in the most efficient way and set it to do so as a default. The question is, how do we get customers to leave it [running] that way," said Frey, who spoke to ZDNET on the sidelines of HPE Discover 2024.

He noted that, most times, businesses operate their data center and IT infrastructures at 30% utilization, choosing to over-provision to ease their anxiety about these systems continuing to operate smoothly.

These businesses end up with a hardware stack that is hyper-efficient, but used inefficiently, he said. He added that a huge part of HPE's efforts go toward helping customers use its products more efficiently, which would in turn reduce the energy needed to power these systems.

Customer education and a change in mindset play a big part in driving overall sustainability efforts, he said. For its part, HPE provides whitepapers and case studies, including adoption frameworks to guide customers through the change, according to Frey.

He noted that metrics and analytics also have a role in quantifying the returns for businesses, be it in terms of dollars, risk reduction, enhanced resilience, reduced carbon emission, or cybersecurity benefits.

Also: Singapore keeping its eye on data centers and data models as AI adoption grows

And while regulations and mandating some of these operating standards can help drive adoption, these should be rolled out in collaboration with the industry and user community, he said. This will ensure there are no unintended consequences, such as poorer performance in other areas.

Policies that lead to such unintended results may compel companies to move workloads out of the regulating country, which is not what the government wants in setting out these mandates, he said.

Asked about key barriers to building more sustainable data centers, Frey noted that the one-size-fits-all strategy no longer works, particularly amid the anticipated spike in artificial intelligence (AI) workloads.

Facilities that power AI applications will likely need liquid cooling to maintain or improve energy efficiency for these compute-intensive environments. On the other hand, tapping ambient air may be sufficient to cool the insides of data centers running more general-purpose applications, he explained.

Also: Business sustainability ambitions are hindered by these four big obstacles

Efforts to address higher temperatures, such as Singapore's data center operating standards for tropical climates, also are better suited for traditional workloads, he said.

As companies move toward higher rack power density with their adoption of AI, they will likely need to move to liquid cooling environments, he noted.

Eventually, Frey believes, most data center operators will move in the same direction, as newer more powerful processors generate more heat since they are capable of handling more tasks.

The average IT rack used to run at between 3 and 5 kilowatts (kW) and this has been growing in the past decade to more than 20kW for mainstream computing workloads, he noted.

Power requirements go up further with racks that run AI workloads or train models, hitting more than 50kW per rack. Ambient air alone then will not be sufficient to cool such environments and will drive the need for liquid cooling, he said.

Also: Global tech spending expected to keep climbing on AI demand

In fact, demand for liquid cooling has driven the data center thermal management market to $7.67 billion, according to tech research and advisory firm Omdia. It is expected to climb at a compound annual growth rate of 18.4% until 2028, fueled by the adoption and development of AI.

In particular, liquid cooling saw significant growth in China and North America, Omdia said. "The data center thermal management is advancing due to AI's growing influence and sustainability requirements," the research noted. "Despite strong growth prospects, the industry faces challenges with supply chain constraints in liquid cooling and embracing sustainable practices."

The firm added that the integration of AI-optimized cooling systems, strategic vendor partnerships, and ongoing push for energy-efficient and environmentally friendly solutions will shape the industry's development.

"Data center cooling is projected to be a $16.8 billion market by 2028, fueled by digitalization, high power capacity demand, and a shift toward eco-friendly infrastructure, with liquid cooling emerging as the biggest technology in the sector," said Shen Wang, Omdia's principal analyst.

Artificial Intelligence

In a recent post on X, analyst and author Dion Hinchcliffe, mentioned that in enterprise-grade AI, retrieval augmented generation (RAG) integrates database data with generative LLMs to produce highly relevant and contextually rich responses.

This method enhances the depth and accuracy of AI outputs, resulting in more precise and insightful responses by combining the strengths of both databases and LLMs.

Echoing the same was a post by LlamaIndex CEO & co-founder Jerry Liu: “A lot of enterprise developers are creating GPT-like platforms for internal users, allowing them to customise the agent to their specific needs through an easy-to-use interface.

“RAGApp is the most comprehensive open-source project available for launching a RAG/agent chatbot on any infrastructure, all without writing a single line of application code—perfect for end users!”

Why Enterprises are Choosing to ‘RAG’

Hallucinations, rising compute costs, limited resources, and the need for a constant flow of current and dynamic information, have all proven problematic for enterprise-grade LLM solutions.

A user on a Reddit thread said that a key aspect is that fine tuning is a feasible option only if the data source you want to include is finished – RAG can provide information based on changes made yesterday or a document that was created this morning, a fine-tuned system won’t.

Moreover, the most important reasons for enterprises to RAG is the reduction of hallucination and provide more accurate, relevant, and trustworthy outputs while maintaining control over the information sources and enabling customisation to their specific needs and domains.

If we look into techniques to reduce the LLM hallucination via RAG, recently ServiceNow reduced hallucinations in structured outputs through RAG, enhancing LLM performance and enabling out-of-domain generalisation while minimising resource usage.

The technique involves a RAG system, which retrieves relevant JSON objects from external knowledge bases before generating text. This ensures the generation process is grounded in accurate and relevant data.

Many companies aim to integrate LLMs with their data to create RAG applications. For example, Cohere, a leading AI platform for enterprises, a few months ago, introduced command R+, a powerful RAG-optimised LLM for enterprise AI.

RAG or Fine Tune?

Giving an example of the pricing of GPT-3.5, ML engineer and teacher Santiago said, “99% of use cases need RAG, not fine-tuning,” as fine-tuning of the model is more expensive, and both are for different purposes.

On the other hand, fine-tuning enhances LLM capabilities for various applications. It improves sentiment analysis by better understanding text tone and emotion, aiding accurate customer feedback analysis. Additionally, it enables LLMs to identify specialised entities in domain-specific texts, improving data structuring.

However, considering both RAG and fine-tuning, recently a post on LinkedIn by Armand Ruiz, VP of product, IBM, said that fine-tuning and RAG are complementary LLM enhancement techniques.

Fine-tuning adapts the model’s core knowledge for specific domains, improving performance and cost-efficiency, while RAG injects up-to-date information during inference. A recommended approach is to start with RAG for quick testing, then combine it with fine-tuning to optimise performance and cost, leveraging the strengths of both methods for efficient, accurate AI solutions.

He also mentioned, “The answer to RAG vs fine-tuning is not an either/or choice.”

Considerations for choosing between RAG and fine-tuning include dynamic vs static performance, architecture, training data, model customisation, hallucinations, accuracy, transparency, cost, and complexity.

Hybrid models, blending the strengths of both methodologies, could pave the way for future advancements.

Does RAG Happen for Every Enterprise

A recent research paper by Tilmann Bruckhaus shows that implementing RAG effectively in enterprises, especially compliance-regulated industries like healthcare and finance can be tricky. It faces unique challenges around data security, output accuracy, scalability, and integration with existing systems, among others.

“Each enterprise has unique data schemas, taxonomies, and domain-specific terminology that RAG systems must adapt to for accurate retrieval and generation,” noted Bruckhaus, touching upon customisation and domain adoption.

However, the paper also mentioned, addressing these challenges may benefit from techniques in semantic search, information retrieval, and neural architectures, as well as careful system design and integration.

What’s Next

As compared to other methods and techniques, RAG seems to be very useful for enterprises, however, compliance-related industries face some challenges to incorporate RAG, but that can be tackled through semantic search techniques, hybrid query strategies for optimising retrieval, experimental evaluation, and so on.

RAG is advancing day-by-day and will continue to improve, eventually becoming more beneficial for enterprises.

The post Why Enterprises Like to RAG appeared first on Analytics India Magazine.

Growing geopolitical tensions, coupled with supply chain constraints have pushed many countries, including India, to consider building their own semiconductor capabilities. Over time, the US has attempted to hinder China’s semiconductor capabilities and also limit their access to high end Graphics Processing Units (GPUs), an important tool used for AI workloads.

“I really think chips are the new oil. While oil defined the 20th century, chips will define the 21st century and every country will want to own elements of manufacturing,” Krishna Rangasayee, the CEO and founder of SiMa.ai told AIM in an exclusive interview.

Founded in 2018, the startup positions itself as a software company which also builds its own silicon. It focuses on developing AI solutions for the embedded edge computing market.

Despite being headquartered in San Jose, California, SiMa.ai has a significant presence in India, and around half of the startup’s innovation happens in India.

Interestingly, the company has chosen India as a strategic market over China, which is the largest semiconductor market in the world.

Engaging with India, Not China

Rangasayee’s recent visit marked his third trip to India in a year. During these visits, Rangasayee, who has accumulated over 18 years at Xilinx, dedicates much of his time engaging with a diverse array of customers, including automobile manufacturers, drone companies, robotics firms, and public sector entities.

The company is betting on India’s renewed focus on manufacturing and semiconductor fabrication ambitions. “I think India is going to be an AI product focussed leader globally and a lot of innovation in AI will come from India,” he said.

However, for strategic reasons, the company has chosen not to sell its chips in China.

“The world of AI and chips and geopolitics is going to be complicated. We are a startup and we are very focussed on the applications that are coming together in the US and Europe. I have no doubt that it will be happening in India as well,” said Rangasayee.

It’s a bold move for a startup, especially considering China’s status as the world’s largest semiconductor market. In 2022, global semiconductor sales totaled $574 billion, with China capturing a significant $180 billion, equating to a 31.4% market share.

Rangasayee’s decision not to engage with China might also be influenced by the US restrictions that prevent US companies from directly selling chips to China. This was done to limit China’s access to high-end graphics processing units (GPUs) developed by NVIDIA.

On the contrary, China has committed to fulfilling nearly 70% of its domestic silicon requirements internally by 2025. Rangasayee also said his decision to not engage with China has to do with AI regulation.

“AI is a complicated area. At some point, there will be an interesting set of regulations that will come together and as a startup, we need to be careful. We are not a big company and we cannot afford to engage and then revisit our engagement,” he said.

“We have so many opportunities in the US, Europe, India, Korea and Japan and that’s a very large market footprint for us,” he added.

Preparing for a Multimodal World with Second-Gen Chips

The startup also recently announced its second-generation chips, which are fabricated by the Taiwanese semiconductor giant TSMC, are under preview and will be publicly launched in the first quarter of 2025.

Built on 6 nm nodes, these chips are designed to run generative AI models on the edge along with computer vision models. The initial generation of these chips is tailored for computer vision applications at the edge only.

“The gen-one chips were on a 16 nm technology and no doubt we will have improvements in terms of power primarily but also on performance. Our partners continue to be TSMC, Arm and Synopsis and we are really appreciative of the help they provide in building this platform,” he said.

The startup aims to build one platform catering to original equipment manufacturers (OEMs) running convolutional neural networks (CNNs), vision transformers, as well as large multimodal models (LMMs).

Rangasayee believes multimodal is the future and this is what the startup has been preparing for.

“Multimodal is going to be everywhere, from every device to appliances, be it a robot or an AI PC. You will be able to converse, watch videos, parse inputs, just like you talk to a human being.”

The startup has also outlined plans for a third generation of chips. Rangasayee mentioned the company’s potential to sell its chips to phone and PC makers but emphasised a cautious approach. However, the startup has no intention of developing chips for data centres.

AGI Might Not Happen in Our Lifetime

While Rangasayee envisions an AI-dominated future, we did ask his thoughts on the current discourse on artificial general intelligence (AGI).

While some notable figures have predicted AI to supersede human-level intelligence within this decade, Rangasayee has a different outlook.

He believes the current AI systems are nowhere near sentience.“I don’t see that happening with the current form of AI, at least in the next 50-100 years. I don’t see a rationale for how that’s connected.

“A lot has been written about it but what is happening is we are moving into an element of multimodality and mimicking things that really help us but the underlying architectures are far (x3) away from being sentient.”

Moreover, many in the industry hold the media complicit in hyping the capabilities of the current AI systems. Rangasayee agrees, “Many smart minds have already spoken about it, but unfortunately the media hypes the capabilities of AI.

Nonetheless, he also adds that “You can never say never. I am an engineer at heart, and I live this every day. I don’t see a roadmap for AI systems having better human capacity and capability.”

The post Why This Semiconductor Chip Company Chose India Over China appeared first on Analytics India Magazine.