Virtual Influencer
Highlights
1. Virtual influencers are getting traction in marketing due to its ability to elicit favorable responses, cost-effectiveness, predictability, and customizability.
2. Stanford 2024 AI Index Report notes that AI systems need new benchmarks, now that they beat humans in many tasks.
3. Robots will soon be a lot more useful because of their affordability and multimodal AI. As an example, Mentee Robotics has revealed Menteebot, a prototype for real-world tasks.
4. Microsoft’s VASA-1 model enables users to generate realistic talking faces with an image and an audio file.
5. Meta announced Llama 3, a high-performing open-source LLM while xAI introduced Grok-1.5 Vision.
6. The company Nothing integrates ChatGPT into its earbuds.
7. Thermal batteries that store and dispatch energy in the form of heat are getting popular in the energy industry.
Innovation Insights
1. Should Your Brand Hire a Virtual Influencer? (Harvard Business Review)
Virtual influencers are gaining popularity in digital marketing. They tend to elicit more favorable responses from followers than human influencers, cost less, and allow for the creation of diverse characters. Marketers can cater to a wide range of demographics and interests. Companies are recognizing that virtual influencers are controlled and predictable, unlike human influencers who may become embroiled in controversies.
2. AI now beats humans at basic tasks — new benchmarks are needed (Nature)
The 2024 AI Index Report from Stanford University indicates that AI systems now nearly match or surpass human abilities in tasks like reading comprehension, image classification, and mathematics. Traditional benchmarks for measuring their capabilities are becoming outdated, prompting calls for new assessment methods that evaluate more complex cognitive tasks.
3. Three reasons robots are about to become way more useful (MIT Technology Review)
Robotics is poised for transformative growth due to cheaper hardware and advanced AI, making research more accessible and enhancing robot intelligence. The affordability and availability of robots like Hello Robot’s Stretch, combined with AI-driven software like Google’s RT-2 and multimodal models such as RFM-1 from Covariant, are enabling robots to perform complex tasks with fewer human demonstrations and more adaptability.
4. Multi-agent collaboration (Deeplearning.ai)
Multi-agent collaboration is the latest AI agentic design pattern, wherein complex tasks are broken into subtasks executed by different AI agents, enhancing task performance. This approach leverages multiple calls to a single or multiple large language models (LLMs) to perform specific tasks. It is exemplified in emerging technologies such as AutoGen and ChatDev, which demonstrate that using multiple agents can provide a flexible framework for decomposing and managing complex projects.
5. The AI revolution will be ‘virtualized’ (McKinsey)
The article underscores the power of digital innovation in product development, emphasizing the use of digital twins and simulations that mirror real-world processes to optimize performance and speed up innovation cycles. It highlights the strategic integration of digital models with physical operations, enabling organizations to enhance efficiency, leverage real-time data, and foster a culture of digital literacy.
AI Innovations
1. AI video tools
Adobe is enhancing Premiere Pro with new generative AI video tools under its Firefly suite, aiming to introduce capabilities such as extending video clips, and adding or removing objects using text prompts. The company is exploring third-party integrations with AI models from Runway, Pika Labs, and OpenAI’s Sora (The Verge).
VASA-1 is a new model designed to create realistic talking faces from a single static image and audio speech, enhancing virtual characters with visually appealing affective skills. The model excels in producing synchronized lip movements and incorporates comprehensive facial expressions and head motions, adding authenticity and liveliness (Microsoft).
2. AI models
Reka, an AI startup, has launched Reka Core, a new multimodal language model capable of interpreting various data forms including images, audio, and video. Core quickly reaches competitive levels against industry leaders like OpenAI and Google, available through API, on-premise, or on-device, enhancing its accessibility (VentureBeat).
3. Chatbot
Poe is quickly evolving with the ambition to serve as a central hub for accessing a diverse range of AI chat models, bolstered by a $75 million funding round and innovative features like multi-bot chat, allowing simultaneous interactions with various AI bots in one thread. This platform facilitates dynamic engagements, where users can utilize different AI models for specialized tasks within a single conversation, streamlining the user experience in finding the best AI tools for specific needs (VentureBeat).
Meta AI, an AI chatbot capable of answering questions, composing poetry, and creating images, is now integrated into Instagram (Engadget). It is also accessible via a standalone website at Meta.ai. Instagram is testing a new “Creator A.I.” program that allows influencers to use chatbots to interact with fans via direct messages, reducing the workload of managing large volumes of fan interactions. The chatbots, designed to mimic the influencers’ unique voices, are currently disclosed as A.I.-generated and aim to maintain high engagement with followers while easing the personal response burden on the influencers (New York Times).
4. API
Stability AI has expanded access to its text-to-image model, Stable Diffusion 3, by making it available to developers through an API and a new platform called Stable Assistant Beta, which is still under preview and not open to the public. The company aims to collaborate with its community to refine the model, which boasts improvements over competitors in areas like typography and prompt adherence (The Verge).
OpenAI’s Assistants API got some updates, such as retrievals for up to 10,000 files and tool choice (OpenAI).
5. Professional assistant
Thomson Reuters announced the expansion of CoCounsel, its professional-grade GenAI assistant, which will integrate Thomson Reuters’ diverse product capabilities into a unified user experience across sectors like Legal, Tax, Risk & Fraud, and Media (Thomson Reuters).
6. Drug discovery
Researchers at the University of Cambridge have significantly accelerated the search for treatments for Parkinson’s disease using AI. The team developed a machine learning strategy to identify compounds that prevent the aggregation of alpha-synuclein, a protein linked to Parkinson’s, from a chemical library of millions of entries. This approach has enabled the identification of several potent compounds and promises to expedite the development of new therapies (University of Cambridge).
7. Slack AI
Salesforce has announced the expansion of Slack AI, a paid add-on now available to all paid Slack users, which integrates generative AI across the platform to enhance user interaction by streamlining message management, enhancing search functions, and summarizing conversations (Mashable).
8. Earbuds
Nothing has announced an integration of ChatGPT into its earbuds, enabling users to access the AI directly via a pinch gesture when paired with a Nothing smartphone equipped with the latest Nothing OS. This new feature is supported by devices like the Nothing Ear and Ear (a) (The Verge).
9. Predicting treatment outcomes
Causal machine learning can be used with clinical trial data or real-world data for predicting efficacy and toxicity of treatments. Causal ML allows estimating individualized treatment effects, which may enable personalized clinical decisions (Nature Medicine).
Other Innovations
1. Thermal batteries
Thermal batteries, a technology enabling the storage and dispatch of energy in the form of heat, are gaining traction as a cleaner option to meet the significant global energy demand for industrial heat. These systems convert electricity—often from renewable sources like wind and solar—into heat, storing it in materials such as bricks or molten salt to be used later, thereby potentially reducing industrial emissions significantly (MIT Technology Review).
2. Audio social media
Airchat, a social media app focuses on audio interactions by allowing users to post and respond with voice recordings instead of text, which are then transcribed within the app. Airchat offers a user-friendly interface where audio plays automatically, aiming to enhance social dynamics and connectivity by using the natural human voice (TechCrunch).
3. Robotics
Boston Dynamics has introduced a new, fully electric version of its humanoid robot, Atlas (video), which is specifically designed for real-world applications. The development of this new Atlas model aims to enhance automotive manufacturing capabilities and promises greater strength and range of motion (NBC News). Mentee Robotics has revealed Menteebot, a prototype humanoid robot that incorporates advanced computer vision and generative AI to potentially perform household and industrial tasks. The robot is designed to understand and execute tasks using LLMs, focusing on locomotion and dexterity (TechCrunch).
4. Neuromorphic computer
Intel has unveiled Hala Point, the world’s largest neuromorphic computer featuring 1.15 billion neurons, aimed at achieving brain-scale computing capabilities. This groundbreaking system is designed to handle complex optimization problems with significant improvements in computational speed and energy efficiency, potentially revolutionizing areas such as drug development (ZDNet).
