- NATURAL 20
- OpenAI Adds Watermarks to Images
OpenAI Adds Watermarks to Images
PLUS: Microsoft Copilot AI to OneDrive, Nvidia and Cisco Partnership and more…
OpenAI is adding new watermarks to DALL-E 3
OpenAI is updating its image generator, DALL-E 3, by adding watermarks to image metadata. These watermarks, following standards from the Coalition for Content Provenance and Authenticity (C2PA), will be visible in images generated on ChatGPT and the DALL-E 3 API. Mobile users will see them by February 12th, featuring both invisible metadata and a visible CR symbol.
The watermarks aim to identify the AI tool used to create the content but are currently limited to still images, not videos or text. OpenAI assures that this addition will have minimal impact on latency and image quality. However, watermarks can be easily removed.
Meta to add ‘AI generated’ label to images created with OpenAI, Midjourney and other tools
Meta, the company behind Facebook and Instagram, plans to label AI-generated images on its platforms, including those created by third-party tools like Google and Adobe. The move aims to address concerns about the spread of misleading information ahead of the 2024 elections.
Meta will also work with other firms to establish industry standards for identifying AI-generated images. The labels will roll out in multiple languages and apply to Facebook, Instagram, and Threads. The company is also developing features to allow users to identify AI-generated video or audio content and prevent the removal of invisible watermarks from such images.
Cisco and NVIDIA to Help Enterprises Quickly and Easily Deploy and Manage Secure AI Infrastructure
Cisco and NVIDIA are collaborating to provide enterprises with simplified cloud-based and on-premises AI infrastructure, networking, and software solutions. This partnership aims to facilitate the deployment and management of secure AI infrastructure, offering end-to-end AI solutions and access to NVIDIA AI Enterprise software for advanced AI and generative AI workloads.
The Ethernet networking-based solutions will be available through Cisco's global channel, enhancing AI cluster deployment. Key customers like ClusterPower have already joined this effort to enhance data center operations with AI/ML solutions. This collaboration aims to enable enterprises to harness the power of AI more effectively.
LinkedIn has rolled out a new AI feature for Premium subscribers. This tool uses your profile info to assist you in starting conversations. When you're about to message someone, the AI jumps in to craft an introduction or inquire about their work experience at a company. It's designed to make networking smoother.
Last year, LinkedIn introduced a similar tool for recruiters and added more perks for Premium users, like job hunting assistance and post summarization. So, if you're a Premium member, this AI buddy might make your LinkedIn game even stronger.
Entrust is buying AI-based ID verification startup Onfido, sources say for more than $400M
London-based startup Onfido, known for using AI in identity verification, is being acquired by Entrust for over $400 million. Entrust, a provider of certification and verification services, plans to integrate Onfido's AI tools into its technology stack, aiming to bolster its position in the identity verification market.
Onfido, founded in 2012, gained traction during the COVID-19 pandemic as the demand for digital identity verification surged. The acquisition reflects the growing importance of identity security and certification technologies in an era of increased cybersecurity threats and data protection regulations.
Entrust, with a strong financial foundation, sees AI as a crucial asset in this landscape, moving towards consolidating various solutions into a more streamlined platform.
Ambience Healthcare raises $70M for its AI assistant led by OpenAI and Kleiner Perkins
Ambience Healthcare, a healthcare startup, just secured a hefty $70 million in funding. They've developed a sort of healthcare operating system to help doctors deal with all the pesky paperwork. They're mainly focusing on various outpatient specialties like cardiology and pediatrics.
While they haven't spilled the beans on how many customers they have, some of the big names in their client list include UCSF, Memorial Hermann Health System, and John Muir Health. Kleiner Perkins and OpenAI’s Startup Fund led this Series B funding round. The total they've raised is now $100 million, and they're on a mission to make doctors' lives easier with AI-powered admin support.
DeepSeekMath 7B is a model trained on 120 billion math-related tokens, which scored 51.7% on the MATH benchmark, rivaling Gemini-Ultra and GPT-4, without extra tools. It's powered by a smart web data use and a new optimization technique, improving math reasoning and memory efficiency.
ConsiStory is a new approach to generate consistent images from text without training. It uses shared internal activations in a pre-trained model, incorporating a unique attention block and feature injection for consistency across different prompts. It achieves top-notch results in keeping subjects consistent and aligned with texts, and it can handle multiple subjects without any optimization steps, marking a significant advancement in text-to-image generation tech.
The paper discusses the growing interest in making machine learning, especially large language models (LLMs), understandable. As LLMs become better at tasks, they also offer new ways to explain complex patterns in simple terms. Yet, challenges like fake explanations and high computational costs arise. The authors suggest focusing on two main areas: using LLMs to analyze new datasets and creating interactive explanations, aiming to enhance our ability to audit and understand LLMs' decisions and outputs.
The paper presents a technique called "Shortened LLaMA," focusing on depth pruning of large language models (LLMs) to reduce their computational demands without sacrificing performance. Unlike width pruning, which trims model dimensions, depth pruning eliminates entire layers, offering a more efficient way to run LLMs under memory constraints. This method promises faster inference speeds and potential deployment on local and edge devices, showcasing an innovative approach to making LLMs more accessible and practical for various applications.
DynamiCrafter is a new approach to make still images move by using video diffusion methods. It makes animations more natural and logical by using a text-to-video model that takes an image as a guide. This tech could bring any picture to life with realistic movements, surpassing older methods limited to specific types of motion. It's a big step forward in animation tech, promising more lifelike and varied animations from regular photos.
Fabric - open-source framework for augmenting humans using AI.
Deepgram - a conversational text-to-speech API for voice AI agents. It aims to deliver human-like conversation quality at high speed and efficiency.
Conveyor - GPT-powered automation software simplifies security questionnaires with over 90% accuracy, saving infosec teams significant time.
Fathom - free AI Meeting Assistant that records, transcribes, and summarizes meetings, supporting multiple languages.
Cape Privacy - a privacy-focused Conversational AI platform with various AI models, including GPT-3.5 Turbo and GPT-4.
MagicBot, More Than A Helper.
You deserve to be free from endless chores.
— MagicLab (@MagicLab244144)
Feb 2, 2024
Two Texas companies were behind the AI Joe Biden robocalls
Lingo Telecom and Life Corporation, both Texas-based companies, were identified as behind the AI Joe Biden robocalls in New Hampshire. The calls featured a deepfake message urging Democratic voters not to participate in the primary election. Cease-and-desist orders and subpoenas have been issued to these companies. THE VERGE
Ikea’s AI assistant gives design inspiration — at least it tries to
Ikea's AI assistant, powered by GPT, aims to simplify the shopping experience, but it falls short of expectations. Users reported that instead of providing design ideas, the AI directs them to the Ikea website. It struggled to answer specific questions about product availability and design preferences. THE VERGE
Colossyan uses GenAI to create corporate training videos
Colossyan, a company co-founded by Dominik Mate Kovacs, aims to make corporate training videos more engaging using generative AI. They utilize AI to generate workplace learning videos, allowing users to input scripts and choose from various avatars. Colossyan's text-to-speech (TTS) engine can read the script in over 70 languages. Colossyan has attracted customers like Novartis, Porsche, Vodafone, HPE, and Paramount, and it recently raised $22 million in funding to expand its capabilities and presence. TECHCRUNCH
Microsoft's AI engine powering H&R Block's tax filing assistance product
H&R Block has partnered with Microsoft to create an AI-powered tax filing assistance tool called AI Tax Assist. This product was developed in collaboration with Microsoft's Azure OpenAI Service and was launched ahead of the tax season. AI Tax Assist uses data from The Tax Institute at H&R Block and insights from over 60,000 tax professionals to assist individuals, self-employed individuals, and small business owners in preparing their tax returns. The tool provides information on tax forms, deductions, credits, and can answer tax-related questions, as well as guide users through the tax filing process. It aims to maximize refunds and minimize tax liabilities. FOX BUSINESS
Elon Musk's AI startup is courting potential investors with a pitch deck that touts access to the 'Muskonomy'
Elon Musk's AI startup, xAI, is attracting potential investors by highlighting its unique selling point: access to what it calls the "Muskonomy." The company's pitch deck, as seen by Bloomberg, emphasizes its connection to Musk's network of companies, including Tesla and SpaceX. The deck mentions Musk's track record as an entrepreneur, especially in the field of deep tech and new frontier businesses that require significant engineering expertise. xAI also cited the success of its rival, OpenAI, as a factor that should appeal to investors. BUSINESS INSIDER
Microsoft to add Copilot AI to OneDrive, but it will cost you
Starting in May, Microsoft 365 users will have the ability to utilize the AI-powered Copilot tool to analyze their OneDrive files. This feature will enable users to ask questions and seek information about their online files without having to open them. Copilot in OneDrive will support various file types and will require a Microsoft Copilot for Microsoft 365 license. Copilot was launched in November and offers AI-powered assistance in various Microsoft 365 applications, but it comes with a subscription fee. Users can choose between Copilot Pro and Copilot for Microsoft 365, with the latter being more cost-effective but offering fewer features. ZDNET
AI chatbot shows surprising talent for predicting chemical properties and reactions
A recent study has demonstrated that a machine-learning system, similar to ChatGPT, can be fine-tuned to effectively answer research questions in the field of chemistry. By adapting a general-purpose machine-learning system and providing it with data related to chemical compounds and materials, researchers were able to make it proficient in predicting properties, reactions, and arrangements of molecules and materials. This approach, which reduces the need for specialized machine-learning models, opens up possibilities for chatbot-like systems to become valuable tools in chemistry laboratories. It simplifies access to machine learning for chemists and can assist in predicting chemical properties with surprising accuracy, even for unknown materials. NATURE
Some OpenAI Investors Sit Out Share Sale, Reflecting AI Divide
Some investors in OpenAI, a prominent artificial intelligence (AI) startup, are sitting out its share sale due to concerns over high valuations and increased competition in the AI sector. Founders Fund, a well-known venture capital firm, initially purchased OpenAI shares in January 2023 at a $29 billion valuation but opted not to acquire more shares during OpenAI's recent employee and investor share sale at an $86 billion valuation. This caution reflects the growing skepticism in the investment community regarding the sustainability and competitiveness of AI startups, as well as concerns about the dominance of tech giants like Google and Amazon in the field. THE INFORMATION
No GPU? No problem. localllm lets you develop gen AI apps on local CPUs
In today's fast-paced AI world, developers often struggle with the limited availability of GPUs needed for large language models (LLMs). Google Cloud introduces a solution: "localllm," allowing developers to use LLMs efficiently on CPUs and memory within Cloud Workstations. Quantized models, optimized for local devices, improve performance, reduce memory usage, and speed up inference.
This approach eliminates GPU dependency, offering flexibility, cost-efficiency, and improved data security. "localllm" provides tools to access HuggingFace's quantized models, enhancing productivity and seamless integration with Google Cloud services. Developers can explore the GitHub repository for detailed instructions on setup and usage.
What'd you think of today's edition?
What are MOST interested in learning about AI?
What stories or resources will be most interesting for you to hear about?