<div>
  <p>
  </p>
  <ul>
    <li>
      <p>
        Meta's Fundamental AI Research (FAIR) team has announced the public release of several new AI models and tools for researchers. These include image-to-text and text-to-music generation models, a multi-token prediction model, and a technique for detecting AI-generated speech. The models are being released under various licenses ranging from research-only to commercial.
      </p>
    </li>
    <li>
      <p>
        Chameleon, which was publicly released, is a family of mixed-modal models that can process and generate text and images. The Chameleon 7 billion and 34 billion models, released under a research-only license, can reportedly handle tasks involving visual and textual understanding, such as image captioning. Another model, JASCO, is designed for text-to-music generation and allows users to control aspects like chords, drums, and melodies through text inputs.
      </p>
    </li>
    <li>
      <p>
        A multi-token prediction model for code completion is also being released under a non-commercial, research-only license. Meta is also releasing AudioSeal, an audio watermarking technique for detecting AI-generated speech within longer audio snippets. 
      </p>
    </li>
  </ul>
  <p>
  </p>
</div>


<div>
  <p>
    <a href="https://about.fb.com/news/2024/06/releasing-new-ai-research-models-to-accelerate-innovation-at-scale/">
      Company blog
    </a>
  </p>
</div>


Generative AI Applications

INCUMBENTS_PRESENT

PROMINENT_CATEGORY

<h2>Summary</h2>
<div>
 <p>Meta's Fundamental AI Research (FAIR) team has made several new AI models and tools public for researchers. These comprise models for image-to-text and text-to-music generation, a multi-token prediction model, and a method for recognizing AI-generated speech. These models are made available under various licenses, from research-only use to commercial application.</p>
 <p>Among the released models is Chameleon, a mixed-modal family of models that can generate and process text and image data. The Chameleon 7 billion and 34 billion models, offered under a research-only license, are equipped to manage tasks related to visual and textual comprehension such as image captioning. JASCO, another model designed for converting text to music, allows users to manipulate elements like chords, drums, and melodies through text commands.</p>
 <p>In addition to the above, a multi-token prediction model for code completion is also being made public under a non-commercial, research-only license. Meta is also launching AudioSeal, an audio watermarking technique for identifying AI-created speech within extended audio sequences.</p>
 
</div>

Meta releases new AI research models for multimodal tasks

Autonomous and connected technologies shaping the future of the auto industry

Auto Tech

ANALYST_QUICK_TAKE

<div>
 <p>Qualcomm Technologies and Google are embarking on a multi-year strategic partnership to build a standardized platform for in-car systems. The collaboration will incorporate Qualcomm's Snapdragon Digital Chassis, Google's Android Automotive Operating System (AAOS), and Google Cloud, with a particular focus on utilizing GenAI.</p>
 <p>The aim is to develop engaging map experiences, intuitive voice assistants, and live software updates for vehicles. This approach simplifies the production process for automakers, who can then create AI-enhanced cockpit systems and provide customizable services for their customers. </p>
</div> <p><b>Analyst QuickTake:<br>
</b> The automotive industry is among many to have witnessed an increasing discourse surrounding the possibilities of incorporating GenAI technologies. In fact, automakers and auto part manufacturers, including <a href="https:<br>
//sp-edge.com/updates/26722"> Peugeot, </a> <a href="https:<br>
//sp-edge.com/updates/21380"> General Motors, </a> <a href="https:<br>
//sp-edge.com/updates/19470"> Mercedes-Benz, </a> <a href="https:<br>
//sp-edge.com/updates/26351"> Volkswagen, </a> <a href="https:<br>
//sp-edge.com/updates/21512"> Continental, </a> and <a href="https:<br>
//sp-edge.com/updates/27058"> Bosch </a> , have collaborated with various GenAI tech providers to enhance their offerings with the technology, mostly to develop voice assistant technologies.</p>

Qualcomm and Google partner to develop AI-driven automotive solutions

Empowering Tomorrow's Innovations with Generative AI Infrastructure

Generative AI Infrastructure

<div>
 <p>Meta's research division has unveiled a new tool called LayerSkip that aims to speed up large language models (LLMs). The solution features a distinctive training recipe coupled with self-speculative decoding to minimize LLMs' computational and memory needs.</p>
 <p>LayerSkip incorporates three primary elements: a training recipe implementing layer dropout and early exit loss, an inference method permitting early exits at premature layers, and self-speculative decoding for early detections and adjustments. The mechanism takes advantage of shared weights to bypass layers, ensuring both high-quality results and efficiency.</p>
 <p>Meta AI reports that LayerSkip has demonstrated significant speed enhancements across different Llama model sizes and functions. For instance, it has attained up to 2.16× acceleration on CNN/DM summarization, 1.82× on coding tasks, and 2.0× on the TOPv2 semantic parsing task.</p>
</div> <p><b>Analyst QuickTake:<br>
</b> The news complements Meta's previous <a href="https:<br>
//sp-edge.com/updates/31479"> announcement </a> to implement a multi-token approach for streamlining LLM development and deployment . The prediction models forecast multiple future words simultaneously, claiming better performance and shorter training periods.</p>

Meta AI releases LayerSkip to accelerate inference in LLMs

Additive manufacturing makes objects by systematically adding material via a CAD program

Additive Manufacturing

<div>
 <p>Freeform, a metal 3D printing solutions provider, has secured an undisclosed amount in funding from NVIDIA's NVentures and AE Ventures.</p>
 <p>The company intends to use this funding to increase its portfolio of printable materials and enhance production capabilities for defense, aerospace, energy, semiconductors, and automotive industries.</p>
 
</div> <p><b>Analyst QuickTake:<br>
</b> The metal additive manufacturing lending space has seen a large influx in investor interest within the last three months, with a number of startups raising funding. This includes <a href="https:<br>
//sp-edge.com/companies/511081"> Titomic </a> , <a href="https:<br>
//sp-edge.com/companies/1558694"> Fortius </a> , and <a href="https:<br>
//sp-edge.com/companies/910329"> Amaero </a> , which raised <a href="https:<br>
//sp-edge.com/updates/35052"> USD 20.6 million </a> , <a href="https:<br>
//sp-edge.com/updates/34568"> USD 2 million </a> , and <a href="https:<br>
//sp-edge.com/updates/33656"> USD 16.9 million </a> , respectively. Freeform differentiates itself from its competition by offering a proprietary technology stack that combines advanced sensing, real-time controls, and data-driven machine learning, which enables users to produce digitally verified, faultless parts.</p>

Freeform secures funding from NVIDIA's NVentures

Using data and automation to achieve 24/7 production.

Smart Factory

<div>
 
 <p>Flexxbotics, a firm that provides robotic control solutions, has enhanced its robotic machine tending with compatibility for LMI Technologies' 3D scanning and inspection products. This new integration allows businesses to implement robot-based manufacturing and maintain six sigma consistency in automated operations.</p>
 <p>The solution by Flexxbotics uses its own FlexxCORE technology to achieve a secure connection and communication between robotics and the equipment from LMI Technologies. It is compatible with the full range of Gocator sensors from LMI Technologies, such as 3D point profilers, 3D line profilers, and 3D snapshot sensors, among others.</p>
 <p>According to Flexxbotics, this compatibility enables the making of real-time adjustments to CNC machine programs using automated inspection results, thus facilitating autonomous process control.</p>
 
</div>

Flexxbotics announces compatibility with LMI Technologies for quality inspection

Data Infrastructure & Analytics

<div>

 <p>Oxla, a data warehousing solutions company, has secured USD 11 million in seed funding. The funding round was lead by TQ Ventures, and included contributions from Lead Ventures, Warsaw Equity Group, and 4Growth VC.</p>
 <p>The company intends to utilize the funds for boosting the commercialization and development of its products. Additionally, the funding will assist in expanding the company's market share and catering to the unmet demand in the data warehousing sector.</p>
 <p>Operating from Poland, Oxla specializes in developing an analytical database designed for processing substantial volumes of data. The firm asserts that its database technology boasts analytical query execution speeds that are 10x faster, and costs that can be up to 85% lower than those of its competitors. Oxla primarily provides services to industries pertaining to the Internet of Things (IoT), industrial applications, ecommerce, and cybersecurity.</p>
 
</div>

Oxla raises USD 11 million in seed funding to drive commercialization

<div>
 
 <p>Cohesity, a data management and storage company, has made some changes to its AI-powered search assistant, Cohesity Gaia.</p>
 <p>The fresh visual data exploration feature leverages topic modeling and natural language processing to automatically spot hidden thematic structures across documents and files. It provides a visual picture of data sorted by themes, enabling users to navigate through each theme, inquire conversational questions, and interact with intelligent, context-aware prompts. The update expands support for more data sources including Microsoft 365 Mail, SharePoint, OneDrive, and on-premise or cloud-based file servers.</p>
 <p>The firm affirms that these updates will offer in-depth, contextual insights while adhering to data security and regulatory compliance requirements.</p>
 
</div> <p><b>Analyst QuickTake:<br>
 </b> This news marks the company's ongoing efforts to help customers securely leverage AI for enhanced data insights. Earlier this year, the company <a href="https:<br>
//sp-edge.com/updates/27747"> partnered with Nvidia </a> to deploy advanced GenAI capabilities into its platform, Gaia.</p>

Cohesity enhances Gaia, its AI assistant, with visual data exploration and expanded data sources

Democratizing financial technology with a suite of API-based solutions

FinTech Infrastructure

<div>
 <p>Finzly, a banking-as-a-service (BaaS) provider, has introduced its FedNow service on the Finzly BankOS platform through the AWS Marketplace. This service is integrated with Finzly's digital banking systems, allowing the ability for instant payments. Improvements in speed, quality, and real-time security are claimed benefits of using this platform.</p>
 <p>As a FinTech infrastructure firm, Finzly offers the BankOS platform. This platform enables financial institutions to connect with standard and advanced payment networks such as FedNow and real-time payments (RTP). The platform also includes unique features like ready-to-launch payment rails and on-demand APIs.</p>
</div>

Finzly launches FedNow service through BankOS platform in AWS marketplace

<div>
 <p>Runway, a GenAI startup specializing in video creation, has introduced a new feature known as Act-One. This advanced tool enables users to capture facial expressions using their smartphone cameras, reproducing them on AI-generated video characters. The announcement is specifically targeted at Runway account users who have enough credits for the Gen-3 Alpha video generation model.</p>
 <p>Act-One provides the ability for users to record either themselves or other actors using any type of video camera, including those on smartphones. The technology then transfers the captured facial expressions to AI-generated characters. Exceptional in its capability, the tool can accurately replicate a range of features including micro-expressions, eye-lines, and even the subtleties of pacing. This can be achieved across a wide variety of character designs and styles, eliminating the need for expensive motion capture equipment or time-consuming manual face rigging.</p>
</div> <p><b>Analyst QuickTake:<br>
 </b> This launch represents a notable advancement in its product offerings. This new feature makes animation accessible to a wider range of creators, regardless of their experience level. As Runway continues to innovate, it is likely to strengthen its position in the AI video space amid the emergence of several new companies launching video models.&nbsp;</p>

Runway launches Act-One for AI facial expression motion capture

<div>
 <p>A Canadian AI image startup, Ideogram, has launched a new product called Ideogram Canvas. This innovative platform serves as an endless creative board for arranging, generating, editing, and merging images. It's accessible to all Ideogram users, but premium versions offer additional features and fewer restrictions.</p>
 <p>The Ideogram Canvas enables users to either upload their personal images or generate fresh ones. The platform's key features include Magic Fill and Extend tools, which offer a wide range of editing options. Magic Fill allows specific image sections to be edited, permitting users to substitute objects, insert text, or modify backgrounds. The Extend tool, on the other hand, enables images to be expanded beyond their original limits while preserving the existing style.</p>
</div>
 <p><b>Analyst QuickTake:<br>
</b> Ideogram raised <a href="https:<br>
//sp-edge.com/updates/27082"> USD 80 million </a> in a funding round earlier this year along with the launch of the image generation model "Ideogram 1.0." In April, it <a href="https:<br>
//sp-edge.com/updates/28467"> updated </a> the model with new capabilities, including description-based referencing and negative prompting, to reportedly enhance the quality and coherence of outputs.</p>

Ideogram launches Canvas for image manipulation and generation

<div>
 <p>UiPath, an AI software firm, has collaborated with Inflection AI for improved enterprise efficiency and security. The partnership aims at integrating Inflection AI within UiPath Autopilot and offering built-in integrations to support private cloud solutions for agentic automation. This caters to industries with high security needs.</p>
 <p>As a result of Inflection AI's alliance with Intel, UiPath will be available as an option in Intel's Tiber AI Cloud service, using the new Gaudi 3 processors. This will enable UiPath customers to keep their data on-site while also utilizing Inflection AI's system and agentic automation.</p>
</div>

All Updates

Meta releases new AI research models for multimodal tasks

Qualcomm and Google partner to develop AI-driven automotive solutions

Meta AI releases LayerSkip to accelerate inference in LLMs

Freeform secures funding from NVIDIA's NVentures

Flexxbotics announces compatibility with LMI Technologies for quality inspection

Oxla raises USD 11 million in seed funding to drive commercialization

Cohesity enhances Gaia, its AI assistant, with visual data exploration and expanded data sources

Finzly launches FedNow service through BankOS platform in AWS marketplace

Runway launches Act-One for AI facial expression motion capture

Ideogram launches Canvas for image manipulation and generation

UiPath partners with Inflection AI to integrate AI solutions for enterprises

Generative AI Applications

Meta releases new AI research models for multimodal tasks

Contact us