Alpha3D wants to accelerate digital asset and AI-powered generation through cheaper, hyper-scaling technology

Alpha3D wants to accelerate digital asset and AI-powered generation through cheaper, hyper-scaling technology Jacquelyn Melinek 7 hours

In the artificial intelligence world, there’s no shortage of image or text generators that allow users to spur up content with the push of a button. Alpha3D, a generative AI-powered platform, is participating in the TechCrunch Disrupt 2023 a participant of the Startup Battlefield 200 cohort in hopes of showcasing a technology that can hyper-scale digital assets.

Its platform aims to help anyone — regardless of 3D modeling knowledge — to generate 3D digital assets from text prompts or uploaded images within minutes. The digital assets can then be used in augmented reality (AR), virtual reality (VR), virtual try-on (VTO), gaming, the metaverse or as an NFT.

The company’s proprietary AI technology was developed internally and has gathered millions of data assets over the past years to scale their platform, Madis Alesmaa, CEO of Alpha3D, said.

Instead of scanning or manual modeling content, Alpha3D’s AI takes an existing 2D image, reads the information from it and constructs a 3D image within a few seconds, he added. Customers can then use that asset in their own gaming environments, metaverses, e-commerce platforms or “wherever they want.”

Its co-founders include Alesmaa, Rait-Eino Laarmann, Shahab Anbarjafari and Mariliis Retter. Alesmaa and Laarmann have backgrounds in building startups, Anbarjafari was a team lead for photogrammetry-based 3D model creation at Rakuten and Retter has prior experience in executive marketing roles.

To date, Alpha3D has raised close to $2 million. Investors include venture capitalists and angel investors like Sebastien Borget, co-founder and COO of The Sandbox; Taavi Rõivas, former prime minister of Estonia; and others from Meta, Slack and PwC, to name a few, according to its deck.

As the hype of the metaverse has seemingly slowed down, many big brands are still experimenting with digital collectibles, or NFTs, among other virtual products, in an effort to meet consumers in a new way. Last year, Nike and Starbucks both launched virtual marketplaces and NFT reward loyalty programs, respectively, in an effort to connect with consumers in a new way.

“What triggered us was the content side,” Alesmaa said. “We felt like everyone was trying to do something with AI visualization and that was the biggest bottleneck for the biggest brands, gaming companies.”

So it’s not a huge surprise that demand for 3D-generated content is still prevalent. Not every big business can (or wants to) create an in-house team to build digital assets, as it’s costly, technical and time-consuming, which is where a service like Alpha3D’s could comes in.

“Whether we like it or not this is where the world is going,” Alesmaa said. “Younger generations are spending so much time inside immersive experiences. It’s going to be another world built on top of our existing world and you need a way to create environments and assets easily.”

The startup’s technology provides users a 3D asset in four to five seconds, compared to some platforms that take a few days to generate, Alesmaa said. The cost is also nominal with the first 50 assets free, but models after that are “max one euro” per model, as the company is not focusing on monetization until later this year, he added. “Right now it’s about user acquisition.”

The company launched in mid-March and has close to 100,000 platform users, with a month-over-month growth of about 40 to 45%, Alesmaa shared. It is working with more than 35 companies, including Nvidia, The Sandbox and LVMH. It has 1.4 million contracted ARR and 60,000 MRR, according to its deck.

The next phase of its roadmap will focus on integration and plugging its technology into different infrastructures. “We see ourselves as an infrastructure layer for thousands of businesses out there…They’re struggling with the content side and that’s where we’re stepping in.”

Oracle’s Grand Multicloud Gamble

Oracle is unlike any other hyperscalers (AWS, Microsoft Azure and Google) in the world. At the Oracle CloudWorld 2023 keynote, Oracle chief technology officer Larry Ellison said that customers already have been using multi-cloud products and services, and there are more strong reasons to believe it should be interoperable and interconnected more than ever.

“Last week, I met the CEO of Microsoft, Satya Nadella, and we had a wonderful chat,” said Ellison, saying that there should not be any walls between the clouds – “Cloud Should be Open.”

Citing Salesforce, Microsoft and AWS on how its customers are running their applications on the cloud, Larry looks to eliminate barriers between different cloud platforms and allow them to work together smoothly. “If you want to move your data out of the AWS cloud and put it into a database. It is your data,” avered Larry, hinting that the extended partnership with Microsoft is just the beginning.

Will AWS & Google Cloud Join?

“I think it’s inevitable that we will support other clouds. We’re happy to have that conversation. I would say that it’s not a matter of if; it’s a matter of when it’s going to happen” said Oracle SVP Karan Batta, in an exclusive interaction with AIM, at Oracle CloudWorld 2023, Las Vegas.

“For us multi-cloud is not just one product. It’s a portfolio of products” said Batta, talking about its partnership with Microsoft and the launch of Oracle Database Service for Azure (OracleDB for Azure) last year, and said that they have about 12 regions interconnected with Microsoft so far.

With OracleDB@Azure, customers can now use OracleDB services through Microsoft Azure. Oracle Cloud Infrastructure (OCI) is directly embedded into Microsoft Azure data centres, solving the issue of latency and performance challenges, thereby providing a smooth experience for customers.

“We’ve always thought of OCI as a cloud that should be able to be workable and interoperable within any cloud,” shared Bhatta.

Further, he said that last year Oracle worked on a mechanism where one could run and deploy Amazon’s Aurora through OCI. However, he said that it was more of a proof of concept but it showed the ability, hinting that more cloud providers are likely to partner with Oracle in the coming months.

Why it make sense

“Customers are going to demand it, people want to use the best of breed from intercloud. They want to use the Oracle database from here, and maybe they want to use something else from another cloud provider,” said Batta, emphasising on how enterprises love the multiple-cloud approach.

This makes total sense. For instance, Amazon Bedrock currently hosts models from AI21, Cohere, Anthropic Claude 2, and Stability AI SDXL 1.0. Similarly Google’s Vertex AI’s Model Garden hosts PaLM 2, TII’s Falcon and Meta’s Llama 2 along with several other models. By strategic collaboration with Oracle, their customers would be getting the best of both worlds.

Nadella also stressed: “AI exists because of data. When I look at anything that you do around AI, you need to have access to data,” indicating that there is no better player than Oracle, which focuses on best-in-class enterprise data security, alongside pay as you go (PAYG) and the universal credit annual commitment model among other benefits.

“I personally think at the end of the day customers will end up choosing multiple cloud providers ,” said Batta,“I think it’s a matter of when, if every customer is going to want to deploy multiple clusters.”

Corey Sanders, corporate vice president of Microsoft Cloud, also believes that multi cloud will be the norm. “And part of me hopes that this is the sort of onset to broader partnerships across many of the cloud providers,” he added.

The post Oracle’s Grand Multicloud Gamble appeared first on Analytics India Magazine.

AMD Expands Kria Portfolio with K24 SOM & KD240 Kit

AMD has unveiled the AMD Kria K24 System-on-Module (SOM) and KD240 Drives Starter Kit, expanding its Kria portfolio for industrial edge applications. These innovations aim to streamline development and accelerate time-to-market for motor control and digital signal processing (DSP) solutions.

The Kria K24 SOM is compact, utilising Integrated Fan-Out (InFO) technology to fit into a form factor smaller than a credit card while consuming minimal power. It offers low latency and high determinism, making it ideal for compute-intensive DSP applications in industries such as robotics, power generation, healthcare, transportation, and electric vehicle charging.

Paired with the KD240 Drives Starter Kit, a motor control development platform, these solutions enable rapid deployment without the need for FPGA programming expertise. This simplifies DSP development, making it accessible for entry-level developers.

Hanneke Krekels, Corporate Vice President of Core Vertical Markets at AMD, emphasised, “The Kria K24 SOM delivers high performance-per-watt in a small form factor, perfect for a fast time to market.”

These innovations are set to impact industries heavily reliant on electric motors, which contribute to 70% of global industrial energy consumption. Even small improvements in motor efficiency can yield significant savings.

Greg Needel, CEO of Rev Robotics, praised the Kria SOM portfolio, stating, “We’re able to simplify development, adapt to changing requirements, and build innovative solutions for both commercial and STEM educational customers.”

The Kria K24 SOM is powered by a Zynq™ UltraScale+™ MPSoC device and is supported by the KD240 starter kit. It supports various design flows, including Matlab Simulink and Python, and is compatible with Ubuntu and Docker. AMD’s Vitis™ motor control libraries are also available.

AMD’s expansion into motor control apps simplifies development further. Optional Motor Accessory Packs (MACCP) enhance the experience with additional motor kits.

The Kria K24 SOM is available in commercial and industrial versions, designed for 10-year industrial lifecycles. The commercial version is already shipping, with the industrial version expected to ship in Q4 2023.

In conclusion, AMD’s Kria K24 SOM and KD240 Drives Starter Kit offer efficient, scalable, and accessible solutions for motor control and DSP applications, promising to accelerate innovation in the industrial edge computing landscape.

The post AMD Expands Kria Portfolio with K24 SOM & KD240 Kit appeared first on Analytics India Magazine.

Amazon is refreshing its Fire TV products with generative AI features. Here’s what’s new

People watching TV on the Fire TV app

Amazon's Devices and Services event is in full swing at the company's newest headquarters in Virginia, and some of the biggest updates went straight to the company's Fire TV line. Today, the company announced a new Fire TV Stick 4K Max, a Fire TV Stick 4K, a Fire TV Soundbar, and some generative AI updates to the voice search feature.

With generative AI, users can ask more than ever of Alexa right from their Fire TV or Alexa Remote. The technology makes Alexa requests sound like conversations rather than commands, letting users ask more complex queries than "Find Bluey." Now, users can describe a storyline, make specific requests in different genres, and ask open-ended questions.

Also: Everything Amazon just announced: Alexa updates, new Echo show, Fire tablets, and more

"With a world of content at your fingertips, sometimes the hardest thing to do is decide what to watch. Fire TV has always been great at search, but generative AI takes it to a whole new level," according to Tapas Roy, vice president of Fire TV at Amazon.

"We've leveraged our new large language model to create a more natural and conversational way to find content using natural dialogue. Simply ask Alexa nuanced or even open-ended questions about anything from genres and actors to storylines and scenes, and enter a conversation that will find you something new based on what you like — even if you don't know what you're looking for."

Fire TV Stick 4K Max

The new Fire TV Stick 4K Max is undoubtedly Amazon's most powerful streaming stick. An upgraded 2.0GHz quad-core processor and double the storage as the previous generation, at 16GB, give users a smarter streaming experience with more storage for apps and games.

Amazon's Fire TV Stick 4K Max remains at $60 and continues to support Dolby Vision, HDR, HDR10+, and Dolby Atmos. The new Max also features Wi-Fi 6E support for customers with a compatible router to enjoy little-to-no buffering for a smooth streaming experience.

Also: Every Amazon AI announcement today you'll want to know about

For the first time, Amazon is introducing the Fire TV Ambient Experience to the Fire TV Stick with the new 4K Max model. This experience lets users repurpose the blank TV screen to turn it into a smart display when not in use. With it, users can see artwork or photos, Reminders, Calendars, Sticky Notes for others to see, shortcuts to music applications, and smart home devices like Ring cameras.

Amazon's new AI Art will also be built into the Fire TV Ambient Experience, so users can use Alexa's newfound generative AI powers to generate any artwork they can imagine to become a personalized background for their TVs.

The Fire TV Ambient Experience is already available in Amazon's Omni Series lineup of Fire televisions.

Fire TV Stick 4K

Amazon also announced a new Fire TV Stick 4K with Wi-Fi 6 support and 30% more power than the previous generation, thanks to a 1.7GHz quad-core processor. The new Fire TV Stick 4K remains at $50 and has 4K Ultra HD resolution, support for Dolby Atmos, Dolby Vision, HDR, HLG, HDR10+, and Alexa Home Theater.

Also: Amazon is turning Alexa into a hands-free ChatGPT

Amazon also announced a new partnership to give new Fire TV customers a free six-month subscription to MGM+, included with the purchase of a new Fire TV streaming device or smart TV.

Fire TV Soundbar

Amazon is also launching a new Fire TV Soundbar, poised to be the perfect addition to the Fire TV lineup, supporting DTS Virtual:X and Dolby Audio. The new soundbar, priced at $120, is compatible with all Fire TV devices and has a simple and compact design that is 24 inches long. The Fire TV Soundbar also has Bluetooth for users to connect their phones, tablets, or other streaming devices.

Amazon

OpenAI unveils DALL-E 3, allows artists to opt out of training

OpenAI unveils DALL-E 3, allows artists to opt out of training Kyle Wiggers 7 hours

OpenAI today unveiled an upgraded version of its text-to-image tool, DALL-E, that uses ChatGPT — OpenAI’s viral AI chatbot — to take some of the pain out of prompting.

Most cutting-edge, AI-powered image generation tools today take prompts — descriptions of images — and turn them into artwork in an array of styles, ranging from the photorealistic to fantastical. But crafting the right prompt can be a challenge, so much so that “prompt engineering” is becoming a bona fide profession.

OpenAI’s new tool, DALL-E 3, uses ChatGPT to help fill in prompts. Via ChatGPT, subscribers to OpenAI’s premium ChatGPT plans, ChatGPT Plus and ChatGPT Enterprise, can type in a request for an image and hone it through conversations with the chatbot — receiving the results directly within the chat app. ChatGPT will take a prompt as short as a few words and make it more descriptive, providing more guidance to the DALL-E 3 model.

ChatGPT integration isn’t the only thing that’s new with DALL-E 3. DALL-E 3 also generates higher-quality images that more accurately reflect prompts, OpenAI says — especially when dealing with longer prompts. And it better handles content that’s historically tripped up image-generating models, like text and human hands.

OpenAI DALL-E 3

An image generated by DALL-E 3.

Beyond this, DALL-E 3 has new mechanisms to reduce algorithmic bias and improve safety — or so OpenAI says. For example, DALL-E 3 will reject requests that ask for an image in the style of living artists or portray public figures. And artists can now opt out of having certain — or all of — their artwork used to train future generations of OpenAI text-to-image models. (OpenAI, along with some of its rivals, is facing a lawsuit for allegedly using artists’ copyrighted work to train its generative AI image models.)

The launch of DALL-E 3 comes as the generative AI race heats up, particularly in the image-synthesizing domain. Competitors like Midjourney and Stability AI continue to refine their image-generating models, putting the pressure on OpenAI to stay apace.

OpenAI plans to roll out DALL-E 3 to premium ChatGPT users in October, followed by research labs and its API customers. The company didn’t say when — or whether — it plans to release a web tool.

Google DeepMind Releases AlphaFold Powered AlphaMissense

Since the majority of missense variations found in the human genome are often uncertain, Google DeepMind has introduced AlphaMissense, an adaptation of AlphaFold fine-tuned on human and primate variant population frequency databases to predict missense variant pathogenicity. Missense variants are genetic mutations that change a single nucleotide in a gene, resulting in a different amino acid being incorporated into the protein.

Through the integration of structural context and evolutionary conservation, this model has achieved leading results across a diverse array of genetic and experimental evaluations, all without explicit training on such data. Furthermore, the average pathogenicity score of genes is capable of predicting their cell essentiality, enabling the identification of short essential genes that existing statistical methods struggle to identify. To benefit the scientific community, a comprehensive database of predictions for all conceivable single amino acid substitutions in humans is provided, with 89% of missense variants being classified as either likely benign or likely pathogenic.

The researchers used AlphaMissense to assess all 71m single-letter mutations that could affect human proteins. When they set the program’s precision to 90%, it predicted that 57% of missense mutations were probably harmless and 32% were probably harmful.

How is AlphaMissence different from AlphaFold?

Both AlphaMissense and AlphaFold use deep learning algorithms to make predictions about proteins, but they have different applications. The model was trained on a large dataset of missense variants and their effects on protein function, using in silico methods to predict the effects of these variants. It was tested on a separate dataset of missense variants and compared its performance to other existing methods.

However, AlphaMissense is designed to predict the effects of missense variants on protein function, while AlphaFold is focused on predicting the 3D structure of proteins. While AlphaMissense was trained on a large dataset of missense variants and their effects on protein function, the former was trained on a large dataset of known protein structures.

AlphaMissense has potential applications in personalised medicine and drug development, while AlphaFold has potential applications in drug discovery and understanding protein function.

It outperforms existing “variant effect predictor” software, offering a more efficient means for experts to rapidly identify the mutations responsible for various diseases. Additionally, this program has the potential to detect mutations previously unrecognized in connection to particular disorders, thus assisting medical professionals in prescribing improved treatments.

AlphaFold was the most cited paper in 2022, giving birth to innumerable real-life applications like finding drug for malaria, and Covid-19, delivering gene therapy and much more. So let’s see how AlphaMissense is going to change the life sciences.

Read the full paper here.

Read more: Protein Wars Part 2: It’s OmegaFold vs AlphaFold

The post Google DeepMind Releases AlphaFold Powered AlphaMissense appeared first on Analytics India Magazine.

Feature Store Summit 2023: Practical Strategies for Deploying ML Models in Production Environments

Sponsored Content

Feature Store Summit 2023: Practical Strategies for Deploying ML Models in Production Environments

Hopsworks is organizing the third Feature Store Summit, a free online conference on October 11th, 2023, on how to build production ML systems with a focus on data management for AI.

How is this event relevant to me?

The Feature Store Summit is highly relevant for anyone involved in developing or operating machine learning systems. The speakers will offer practical insights, helping attendees navigate the complexities of productionizing machine learning systems and avoiding common pitfalls. There will be a special emphasis on real-world applications. The summit will help you stay abreast of the latest developments in this rapidly evolving field by sharing experiences and solutions to challenges such as data management, automation and system operation.

What will I learn at the Feature Store Summit?

Feature stores have moved past the top of the "hype cycle", and are now accelerating the production of machine learning models at many organizations.

In this year's talks, we will look beyond just education about what feature stores can do and demonstrate the value feature stores are now creating, boosting feature engineering efficiency, data quality, model reproducibility, and model monitoring. You will learn more about the journey of feature stores from companies such as Hopsworks, Uber, WeChat, Gartner, Databricks and many more. Learn from experts about how to build machine learning systems that deliver real-world value and gain insight into cutting-edge technologies that facilitate the implementation of machine learning models, as well as how to optimize your platform.

Is this event free to attend?

Yes. You will get the chance to connect with other ML professionals and enthusiasts and interact with the speakers directly.

Register for free at the Feature Store Summit 2023

More On This Topic

  • Feature Store Summit 2022: A free conference on Feature Engineering
  • Metadata Store for Production ML!
  • Feature Store as a Foundation for Machine Learning
  • Beyond Brainless AI with a Feature Store
  • ColabCode: Deploying Machine Learning Models From Google Colab
  • A Practical Approach To Feature Engineering In Machine Learning

What’s New in the Latest TensorFlow 2.14

Google has released the latest version of its popular open-source software library, TensorFlow 2.14. The launch comes two months after the introduction of TensorFlow 2.13. For 2023, we already have the third update with the earlier one being released in July.

Check out the GitHub repository to learn more about the update.

Some of the major improvements include a new optional installation method in the TF pip package for Linux that installs necessary Nvidia CUDA libraries. As long as the Nvidia driver is installed on a system, users can now run pip install tensorflow[and-cuda] to install CUDA library dependencies in the Python environment. Aside from the Nvidia driver, no other pre-existing Nvidia CUDA packages are necessary.

Additionally, the team has introduced an API called “strict_mode.” This API serves the purpose of transforming all deprecation warnings into actual runtime errors, complete with clear guidance on transitioning to the suggested replacements. Furthermore, a fresh API, “dtensor.relayout_like,” has been incorporated, to reorganize a tensor according to the layout of another specified tensor.Changes have been made to the functions “tf.ones,” “tf.zeros,” “tf.fill,” “tf.ones_like,” and “tf.zeros_like,” as they now accept an extra “Layout” parameter, for users to control the output layout of the results produced by these functions.

The release also includes support for C++ Run-Time Type Information (RTTI) on mobile and Android platforms. To activate this feature, users should use the following flag when building TensorFlow with Bazel: “–define=tf_force_rtti=true.” This step may become necessary when incorporating TensorFlow into programs that utilize RTTI, as the combination of RTTI and non-RTTI code can potentially lead to compatibility problems at the Application Binary Interface (ABI) level.

TF Lite, Keras and Next Release

For TF lite, users can now add experimental support conversion of models that may be larger than 2GB before buffer deduplication.

The update for Keras, is that model.compile now supports steps_per_execution=’auto’ as a parameter, allowing automatic tuning of steps per execution during Model fit, Model.predict, and Model.evaluate for a significant performance boost.

The updated version will not be henceforth supporting Python 3.8 but TF 2.13.1 patch release will still have the support. After the update, the class hierarchy for tf.Tensor has changed, and there are now explicit EagerTensor and SymbolicTensor classes for eager and tf.function respectively.

Users who relied on the exact type of Tensor will need to update the code to use isinstance(t, tf.Tensor). The tf.is_symbolic_tensor helper added in 2.13 may be used when it is necessary to determine if a value is specifically a symbolic tensor.TensorFlow Debugger CLI: ncurses-based CLI for tfdbg v1 has also been removed.

In the next release, tf.compat.v1.Session.partial_run, tf.compat.v1.Session.partial_run_setup and tf.estimator API will be removed. Users make a note, TF Estimator Python package will no longer be released.

The post What’s New in the Latest TensorFlow 2.14 appeared first on Analytics India Magazine.

10 ChatGPT Projects Cheat Sheet

What You Need to Stand Out

ChatGPT is rapidly changing the game for artificial intelligence capabilities. KDnuggets' latest cheat sheet provides a helpful guide to 10 exciting hands-on projects that demonstrate how to leverage ChatGPT for a variety of data science workflows. From building AI assistants and web applications to generating PowerPoint presentations, the projects outlined offer practical examples across machine learning, natural language processing, and full stack development.

10 ChatGPT Projects Cheat Sheet

The cheat sheet links to tutorials for each project, walking through step-by-step implementation leveraging ChatGPT's conversational prompts. Highlights include using ChatGPT for a loan approval classifier model, resume parser, real-time language translator, exploratory data analysis, and even integrating its capabilities into Google Sheets. Whether you're new to ChatGPT or looking to push its boundaries, this collection of projects acts as a launch pad to boost productivity and accelerate AI-assisted development.

Highlights include using ChatGPT for a loan approval classifier model, resume parser, real-time language translator, exploratory data analysis, and even integrating its capabilities into Google Sheets.

With code examples across Python, React, and more, these carefully chosen projects allow hands-on experimentation with ChatGPT's potential. They provide an ideal way to complement theoretical knowledge with practical experience. As ChatGPT continues its rapid evolution, resources like this cheat sheet offer data scientists an easy reference to unlock its possibilities and integrate AI into their workflows. This is a must-have guide for any practitioner looking to level up their skills with ChatGPT.

Check it out now, and check back soon for more.

More On This Topic

  • The ChatGPT Cheat Sheet
  • Machine Learning with ChatGPT Cheat Sheet
  • ChatGPT for Data Science Cheat Sheet
  • ChatGPT for Data Science Interview Cheat Sheet
  • 10 ChatGPT Plugins for Data Science Cheat Sheet
  • Data Science Cheat Sheet 2.0

Everything we’re expecting at Microsoft’s Surface and AI event this week

Microsoft front of building in NYC

Microsoft's special September 21 event in NYC is around the corner, and like at its previous fall launch events, we are expecting the company to release new Surface products and a whole lot of AI updates.

At last year's Microsoft fall event, the company unveiled the Surface Laptop 5, Surface Pro 9, Surface Studio 2+, and two Surface accessories to optimize the hybrid meeting experience.

Also: How to access thousands of free audiobooks, thanks to Microsoft AI and Project Gutenberg

This year, expect much of the same from the AI leader, with several new Surface product unveilings, including two-in-one models, laptops, and accessories, alongside a healthy dose of AI-powered features and services. At a minimum, Microsoft will likely make previously announced AI features finally available to use.

How are we so sure? Yusuf Mehdi, Corporate VP & Consumer Chief Marketing Officer at Microsoft, basically confirmed via an X post that more AI innovations will be shared during the September 21 event.

Unlike Apple's launch event this past week, Microsoft will not be live-streaming its launches for the general public to tune in to. Instead, expect rolling announcements all day (and week) long, with ZDNET among the press who will be in attendance and reporting from the floor.

Until then, here's a breakdown of all the new hardware and software we're expecting on Thursday.

What's new with hardware?

What's new with software?

More Microsoft