AI News Update | Week 30 Sept-06 Oct 23

Stability AI has recently unveiled an experimental version of Stable LM 3B (which marks an important update from Stable LM released earlier this year), marking the latest addition to their high-performance generative AI solutions. Distinctively, Stable LM 3B, with its 3 billion parameters, is designed to function seamlessly on portable digital devices such as handhelds and laptops.

Features:

1. Compact Design: With 3 billion parameters, Stable LM 3B is considerably more compact compared to the industry’s typical 7 to 70 billion parameters.
2. High Performance: Despite its smaller size, Stable LM 3B surpasses the performance of previous 3B parameter language models and even rivals some of the top open-source 7B parameter models.
3. Versatility: Beyond its primary function as a general language model, Stable LM 3B can be fine-tuned for specialized tasks, including programming assistance.
4. Open-Source Availability: The model is currently available for download on the Hugging Face platform under the open-source CC-By-SA 4.0 license.

Benefits:

1. Efficiency and Accessibility: The smaller size and efficiency of Stable LM 3B mean it requires fewer resources and has lower operating costs, making it accessible to a broader range of users.
2. Eco-Friendly: The model’s design ensures it consumes significantly less power, aligning with environmental sustainability goals.
3. Broadened Application Range: Stable LM 3B’s development expands the scope of viable applications on edge devices or home PCs, allowing for the creation of advanced technologies with robust conversational capabilities.
4. Customizability: Companies can cost-effectively adapt the model to their specific needs, such as customer support or specialized programming assistance.

Other Technical Details:

• Stable LM 3B has demonstrated improved performance in common natural language processing benchmarks, including common sense reasoning and general knowledge tests.
• The model underwent extensive training over multiple epochs using high-quality data, which has contributed to its superior performance compared to its predecessors.
• Developers are advised that Stable LM 3B is a base model, necessitating adjustments for safe performance in particular applications. An instruction fine-tuned model is currently undergoing safety testing, with a release planned in the near future.
• Stability AI envisions that compact, customizable models like Stable LM 3B will increasingly dominate practical generative AI use cases, with open models becoming the standard for trustworthy, auditable AI.

Emerging from stealth mode, Nucleus AI, a startup founded by former professionals from Amazon and Samsung Research, has showcased its capability to develop large language models (LLMs) on par with major tech giants. Their recent launch of a 22-billion-parameter LLM has garnered attention for its performance, surpassing models of similar size.

Features:

• The LLM, available under open-source MIT and commercial licenses, is positioned between 13B and 34B segments.
• Pre-trained on a vast dataset, including web data, Wikipedia, Stack Exchange, arXiv, and coding platforms.
• Future releases include models with 3-billion and 11-billion parameters, trained on larger context lengths of 4,096 tokens. The 22B model will have versions trained on datasets of varying sizes, from one trillion to 350 billion tokens.

Benefits:

• The model’s superior performance promises advancements in AI applications, particularly in agriculture.
• Upcoming models, with different dimensional choices, offer versatility in applications.
• Nucleus AI’s approach aims to make agriculture “failure-proof,” leveraging AI to mitigate potential risks and optimize supply and demand, akin to platforms like Uber.

Other Technical Details:

• The achievement of developing such a model with a team of just four individuals, leveraging the open-source community, speaks volumes about the team’s expertise.
• Nucleus AI joins an exclusive list of companies that have open-sourced and commercially licensed an LLM, setting itself apart as an early-stage startup.
• The company’s focus on agriculture aims to address challenges like optimizing yields. More details on this initiative are expected to be announced in the second week of October.

Conclusion: While many companies are venturing into AI development, Nucleus AI’s focus on agriculture signifies a commitment to tangible real-world impact. The founders, all former founders themselves with rich backgrounds in AI research, are united in their mission to revolutionize agriculture, making it more lucrative and efficient.

• Artifact Introduces AI-Generated Cover Art for Posts

Artifact, the AI-driven news app co-founded by Instagram’s Kevin Systrom and Mike Krieger, has unveiled a new feature allowing users to utilize AI for generating cover images for their content. This announcement comes just a week after the platform introduced the posting capability, further positioning Artifact as a competitor to platforms like Instagram and X (previously known as Twitter).

Originally designed to aggregate news from around the web, Artifact’s evolution now includes direct content sharing and personalization. The latest AI feature, available on both iOS and Android, enables users to craft a post, click the “+” icon, and select “Create with AI.” By inputting descriptions or keywords, users can generate images tailored to their content. For instance, entering “climate” might produce an image of a natural landscape. Additionally, users can define the style, genre, or medium, such as “anime” or “digital art.” If unsatisfied with the generated image, they can adjust their prompts or try new ones.

This AI tool is among several innovative features introduced by Artifact since its January launch, including AI-based article summarization, clickbait detection, and even an option to have news articles read aloud in celebrity voices.

• Zoom Introduces AI-Powered Zoom Docs for Enhanced Collaboration

During the Zoomtopia 2023 event, Zoom unveiled its latest innovation, “Zoom Docs.” This new modular workspace is designed to foster collaboration, and its standout feature is the integration with the Zoom AI Companion. This AI tool aids users in generating content, populating documents, and even summarizing meetings, making the documentation process more streamlined.

Zoom’s move with Docs is a strategic one, positioning the platform as a potential rival to established office suites like Google Workspace and Microsoft 365. Both of these giants have recently incorporated AI tools, but Zoom sets itself apart by including AI capabilities within its standard subscription plans. This approach not only offers enhanced features but also presents a cost-effective solution for businesses.

With a focus on team collaboration and AI-driven functionalities, Zoom Docs is poised to reshape how teams interact and manage content. The broader rollout of this promising tool is eagerly awaited in 2024.

• Microsoft’s Bing Chat Incorporates OpenAI’s DALL-E 3 for Enhanced Image Generation

In a significant move, Microsoft has seamlessly integrated OpenAI‘s advanced image generation model, DALL-E 3, into its Bing Chat platform. This integration, now available to the broader user base, positions Bing a step ahead of OpenAI’s own ChatGPT in terms of DALL-E 3 access.

DALL-E 3 stands out with its refined capabilities, offering users an improved understanding of prompts and the ability to produce images that strike a balance between creativity and realism. Prioritizing user safety, the model is designed with features to prevent the replication of public figures and the generation of inappropriate content. Complementing this, Microsoft ensures that each AI-generated image on Bing Image Creator carries a watermark, serving as a clear identifier.

Beyond Bing Chat, Microsoft envisions a broader application for DALL-E. Plans are in motion to integrate this technology into the Paint app, introducing an innovative AI image creation tool. However, the enthusiastic reception of DALL-E 3 has led to server overloads, indicating the tool’s popularity but also causing temporary access challenges for users.

• Meta’s AI-Generated Sticker Tool Produces Controversial Content

Meta’s newly introduced AI-generated sticker tool for Facebook Messenger has been generating a slew of questionable and often inappropriate stickers. Early user tests have revealed stickers depicting child soldiers, armed Nintendo characters, inappropriate images of iconic figures like Mickey Mouse, and even risqué illustrations of Canadian Prime Minister Justin Trudeau. The tool, which was announced at Meta’s recent Connect event, also showcased a tendency to add inappropriate features to various characters, including Sonic the Hedgehog and Karl Marx.

This sticker tool, powered by Meta’s Llama 2 large language model, allows users to generate unique stickers using text prompts. It is currently being rolled out to select English language users across Facebook Stories, Instagram Stories and DMs, Messenger, and WhatsApp. While certain words and prompts are blocked to prevent the creation of inappropriate content, users have found workarounds using typos or alternative descriptions, leading to the generation of problematic images.

Such challenges are not uncommon with the introduction of AI tools. Meta’s strategy of a limited rollout for the sticker feature is likely to address and rectify these issues before a broader release. Meta has been contacted for comments on these findings.

• Visa Allocates $100 Million for Generative AI Startups

Visa has announced a $100 million venture initiative aimed at investing in startups specializing in generative AI. This endeavor will be spearheaded by the company’s corporate venture division, Visa Ventures. The credit card giant believes that while generative AI has predominantly been utilized for tasks and content creation, its potential to revolutionize commerce is imminent.

Jack Forestell, Visa’s Chief Product and Strategy Officer, emphasized the transformative power of this technology, stating it will not only alter our daily lives and work but will also introduce significant changes to the commerce sector. Since 2007, Visa Ventures has been actively investing in and collaborating with firms dedicated to pioneering advancements in payments and commerce.

• Google Introduces “Assistant with Bard” Powered by Generative AI

Google Assistant, known for its conversational capabilities over the past seven years, is set to undergo a transformative upgrade with the integration of generative AI. This evolution aims to provide users with a more intuitive, intelligent, and personalized digital assistant experience.

Unveiled at the “Made by Google” event, “Assistant with Bard” is a generative AI-powered personal assistant. It merges Bard’s generative and reasoning abilities with the personalized assistance of Google Assistant. This new iteration allows users to interact via text, voice, or images, and it can proactively perform tasks on their behalf. In the upcoming months, “Assistant with Bard” will be available on both Android and iOS devices.

One of the standout features is its integration with popular Google services, such as Gmail and Docs, ensuring users can efficiently manage their tasks and information. On Android, the assistant offers a contextually helpful experience. For instance, after taking a photo, users can overlay “Assistant with Bard” on the image and request a social media post caption. The assistant will then use the image as a reference, understand the context, and generate a suitable caption.

While “Assistant with Bard” is still in its experimental phase, Google plans to release it to a select group of testers for feedback before a broader public launch in the coming months.

• Gates-Backed Startup, Likewise, Launches AI Chatbot “Pix” for Customized Entertainment Recommendations

Likewise, a startup with the backing of tech magnate Bill Gates, has unveiled its innovative AI chatbot, Pix. This advanced tool is designed to curate personalized recommendations for users, spanning books, movies, TV series, and podcasts. Utilizing OpenAI’s cutting-edge natural-language processing technology, Pix is set to redefine the entertainment recommendation landscape.

What distinguishes Pix from other recommendation systems is its ability to harness over 600 million consumer data points. This vast data pool enables Pix to offer suggestions that span across multiple platforms, ensuring users receive diverse and tailored content choices. Whether interacting through text, email, or the dedicated Likewise app, users can expect Pix to refine its suggestions over time, learning and adapting to individual preferences.

Ian Morris, the CEO of Likewise, emphasized the transformative potential of Pix. He stated, “The era of generic recommendations is fading. Personal agent features, like what Pix offers, will be at the forefront of how users discover content in the future.”

In a market flooded with AI tools and chatbots, Pix’s commitment to delivering precise, timely, and cross-platform content sets it apart. The chatbot remains updated with the latest content from popular streaming giants, including Netflix, Hulu, and Max, ensuring users always have fresh and relevant entertainment options at their fingertips.

• IBM Advances PyTorch’s Capabilities Beyond Training to AI Inference

IBM is pushing the boundaries of the open-source PyTorch machine learning framework, traditionally used for AI training, to encompass AI inference. This move is seen as a challenge to dominant players like Nvidia‘s Triton inferencing server. IBM’s objective is not to replace existing technologies but to offer an open-source alternative compatible with various vendor technologies and both GPU and CPU platforms.

PyTorch, initiated by Meta (previously Facebook), transitioned to an open governance model under the PyTorch Foundation at the Linux Foundation in September 2022. IBM, an active member of this foundation, is keen on promoting the enterprise deployment of this open-source tool.

IBM’s approach combines three techniques within PyTorch: graph fusion, kernel optimizations, and parallel tensors, all aimed at enhancing inference speeds. Preliminary results from these optimizations have shown promising inference speeds on large language models. While some of these enhancements will be part of the upcoming PyTorch 2.1 update, IBM plans to contribute more of its inference optimization capabilities to the open-source project. Additionally, IBM is exploring dynamic batching to further scale PyTorch’s inference capabilities for broader enterprise applications.

• SoftBank’s Masayoshi Son Predicts the Advent of Advanced AI Within a Decade

In a significant statement at the SoftBank World 2023 corporate conference in Tokyo, SoftBank’s CEO, Masayoshi Son, shared his vision for the future of artificial intelligence. He expressed a strong belief that artificial general intelligence (AGI) will come to fruition within the next 10 years. AGI, often considered the holy grail of AI research, represents a type of artificial intelligence that can outperform human intelligence across a wide range of tasks and functions.

Son’s prediction underscores the rapid advancements in the AI sector and the potential for machines to achieve cognitive abilities comparable to human intelligence. While the concept of AGI has been a topic of debate and speculation among experts, Son’s statement adds weight to the optimism surrounding its realization.

The implications of achieving AGI are profound, as it could revolutionize various sectors, from healthcare and finance to entertainment and transportation. SoftBank, under Son’s leadership, has been at the forefront of investing in technology startups and innovations, and this prediction aligns with the company’s vision of a future deeply intertwined with advanced AI capabilities.

• LinkedIn Enhances Recruitment with AI-Powered Features

In a significant move to modernize its platform, LinkedIn has unveiled new artificial intelligence capabilities tailored to streamline the recruitment and marketing processes. These AI-driven features are set to revolutionize the way recruiters and HR professionals search for potential candidates. Instead of relying on traditional keyword-based searches, recruiters can now pose questions in natural language, making the search process more intuitive and aligned with human conversation.

Furthermore, LinkedIn’s AI enhancements are not limited to recruitment alone. Marketing professionals stand to gain from an expedited ad campaign creation process. With just a few clicks, they can set up and launch effective campaigns, reducing the time and effort traditionally associated with such tasks.

This initiative underscores LinkedIn’s commitment to staying at the forefront of technological advancements in the professional networking space. By integrating AI into its core services, LinkedIn aims to offer a more user-friendly, efficient, and responsive platform, catering to the evolving needs of its vast user base.

• Move AI Secures $10M for 3D Motion Capture Advancements

UK-based startup, Move AI, has raised $10 million in seed funding to enhance its 3D motion capture application, which converts 2D videos into 3D motion models. This innovative approach offers a cost-effective alternative to traditional motion capture methods used in entertainment. Key investors include Play Ventures, Warner Music Group, and Animoca Brands, highlighting the technology’s potential impact on various entertainment sectors.

Tino Millar, Move AI’s CEO, emphasized the company’s commitment to democratizing 3D animation. Their “Move One” app, currently in beta, is a testament to this vision, aiming to simplify the motion capture process for a wider audience. With this funding, Move AI is set to further its mission, potentially revolutionizing motion capture’s role in the entertainment industry.

• Samsung Partners with AI Startup Tenstorrent for Chip Manufacturing

Samsung’s contract chip manufacturing division has secured a new collaboration with Tenstorrent, a Canadian artificial intelligence chip startup. Announced on Monday, this partnership underscores Samsung‘s continued expansion in the AI chip market. The specific details of the collaboration and the nature of the chips to be manufactured remain undisclosed. The alliance between a tech giant like Samsung and an emerging player like Tenstorrent highlights the growing significance of AI-driven technologies in the chip manufacturing sector.

• Canva Introduces Generative Video Tool and AI-Powered Magic Studio

Canva, the renowned Australian online graphic design platform, has announced a major update to its offerings, introducing a range of AI features. A highlight is the new generative video tool developed in partnership with AI video startup, Runway ML. This tool allows users to input text or a still image, and Canva’s integration with Runway will produce up to 18 seconds of video based on the provided content.

The expansive Magic Studio update brings a suite of AI-driven features. Among them are “Magic Switch,” which transforms a single design into various assets, “Magic Design” that generates unique designs from text prompts or images, and “Brand Voice” which ensures brand consistency across generated content. Other features include “Magic Morph,” “Magic Grab,” and “Magic Expand,” each offering unique design enhancements. Additionally, Canva’s Apps Marketplace will now host selected third-party AI apps.

Addressing security concerns, Canva has introduced “Canva Shield,” an enterprise-grade safety and privacy control system. It offers indemnification to organizations using AI, ensuring content creation is secure and compliant. Furthermore, Canva has launched a “Creator Compensation Program,” a $200 million initiative to compensate creators who allow their content to be used for training Canva’s AI models. The company revealed that its AI adoption has led to significant growth, with 65 million new monthly active users added in the past year. The Magic Studio features are now available to Canva’s paying subscribers.

• Atropos Health Unveils AI System to Enhance Real-World Evidence in Healthcare

Atropos Health, a healthcare startup based in Palo Alto, California, has launched a cutting-edge artificial intelligence (AI) system aimed at transforming the way evidence is gathered to inform medical decisions and research. The company’s Geneva Operating System (OS) employs natural language processing and generative AI to swiftly query real-world clinical data, enabling users without technical expertise to obtain publication-quality results in a fraction of the traditional time.

The system also features a chatbot interface, ChatRWD, where users can ask questions in plain language. The AI suggests query refinements, selects data from Atropos Health’s network of 160 million de-identified patient records, conducts the analysis, and presents the findings in an observational study report. Brigham Hyde, Ph.D., co-founder and CEO of Atropos Health, expressed the company’s ambitious goal: to inform every clinical treatment decision and research inquiry with personalized real-world evidence on a global scale.

Hyde emphasized that Geneva OS and ChatRWD are crafted to offer healthcare professionals swift, consistent, and reliable access to vast real-world data. The platform’s extensive patient records provide a wealth of information. As Atropos Health expands its evidence network and forges new partnerships, it faces challenges in managing and integrating data from diverse sources. Issues of data privacy, security, and the ethical implications of AI in healthcare also arise. However, Hyde remains optimistic about the potential of AI to reshape healthcare, emphasizing Atropos Health’s dedication to expediting evidence generation in the field.

• Reka Debuts Yasa-1, a Multimodal AI Assistant Challenging ChatGPT

Reka, an AI startup with roots in tech giants like DeepMind, Google, Baidu, and Meta, has unveiled Yasa-1, a cutting-edge multimodal AI assistant. Unlike traditional assistants that rely solely on text, Yasa-1 comprehends images, short videos, and audio snippets. Currently in private preview, the system can be tailored to private datasets of any modality, supporting 20 languages and offering internet-sourced contextual answers, long document processing, and code execution.

Yasa-1’s capabilities extend to generating high-level descriptions of multimedia content, though it’s recommended that audio or video clips be under a minute for optimal results. The assistant can also execute code blocks, providing real-time results, and fetch up-to-date content from the web through commercial search engines. However, Reka acknowledges that Yasa-1, like other large language models, can occasionally produce inaccurate outputs and shouldn’t be the sole source for critical advice.

Emerging just three months ago with a substantial $58 million in funding, Reka faces stiff competition from established players like OpenAI and Anthropic. Yet, with a team comprising researchers from major tech companies and a vision to enhance AI’s role in addressing global challenges, Reka is poised to make significant strides in the AI landscape.

• Helios Introduces AI Analyst Cersi for Enhanced Supply Chain Monitoring

Helios, a startup co-founded by former members of Boston Consulting Group and Google, has unveiled Cersi, a conversational AI chatbot tailored for the agricultural supply chain. Unlike traditional chatbots, Cersi specializes in tracking real-time updates related to weather forecasts, natural disasters, climate change effects, and other news events that could disrupt the supply chain. Users can input supplier details, and Cersi will scour the web for relevant information, providing insights in seconds.

The broader Helios platform, of which Cersi is a part, processes billions of climate, economic, and political signals daily, offering a comprehensive view of potential supply chain disruptions. The platform’s dashboard, available to beta customers since March 2023, provides a risk assessment for commodities based on various factors. Cersi enhances this by allowing users to fetch and prepare data using natural language queries.

Francisco Martin-Rayo, Helios CEO, emphasized the company’s mission to prevent food supply shortages and improve supply chain resilience. While Helios aims to support food processors and consumer packaged goods (CPG) companies, commodities traders also utilize the technology. Martin-Rayo believes that even traders can contribute to a more secure food supply chain by being proactive and collaborating with suppliers.

Bogdan Iancu is a seasoned entrepreneur and strategic leader with over 25 years of experience in diverse industrial and commercial fields. His passion for AI, Machine Learning, and Generative AI is underpinned by a deep understanding of advanced calculus, enabling him to leverage these technologies to drive innovation and growth. As a Non-Executive Director, Bogdan brings a wealth of experience and a unique perspective to the boardroom, contributing to robust strategic decisions. With a proven track record of assisting clients worldwide, Bogdan is committed to harnessing the power of AI to transform businesses and create sustainable growth in the digital age.