Data and AI Leader | Coach | Data and AI Strategy Instructor @ Berkeley Haas | ex-Microsoft Data Scientist | Accelerating AI-driven business innovation and leadership transformation
🧠 The Graymatter Review edition summarizes data and AI products, strategies, and innovations to accelerate business value and career growth.
🗣️ ChatGPT Advanced Voice
What
OpenAI released Advanced Voice, a ChatGPT feature that enables more fluid audio conversations. Advanced Voice is multimodal and can produce words and images. The model listens to your voice, emotions, and tone. The upgraded voice capability can respond more quickly and stop talking if you interrupt it.
Key features:
Natural Conversation - real-time, fluid conversions that mimic human language, including the ability to pause if it’s interrupted.
Detecting Emotions - listens to emotional cues in a user’s voice to improve its responses in a more self-awareness tone (e.g., empathy).
Multiple Speakers - handles conversations between speakers by maintaining context in multi-participant meetings or discussions.
Expanded Voices - nine voices to choose from (5 new) and improved accents for foreign languages.
Personalized Experience - you can configure “Custom Instructions” to personalize the voice experience.
Strategy - How to Use It
Here are some ways to derive value from this technology.
Career Growth and Personal Productivity:
Learning a New Language - you can practice over 50 languages to learn how to pronounce and understand words and sentences.
Language Translation - enable real-time conversations with someone in a different language.
Interview Preparation - practice mock interviews and get feedback to strengthen your expertise in a particular subject and soft skills, including communication, confidence, and self-awareness.
Business:
Customer Success - enable 24/7 support for routine customer support cases and requests.
Client Engagement - automate routine and repeatable workflows such as updating clients, scheduling meetings, and personalizing engagement.
Healthcare Patient Experience - automate workflows and improve engagement with patients who may be visually or writing impaired.
Learn More and Try
You must be a ChatGPT Team or Plus user to access the Advanced Voice feature.
Meta launched Llama 3.2 - a collection of open-source AI models in four parameter sizes:
1B and 3B are lightweight, efficient, and multilingual text-only models you can run on mobile and edge devices.
11B and 90B are multimodal models capable of using text and image inputs and outputs text. These are the first Llama models to support vision tasks.
For Developers
These models can be downloaded from llama.com or Hugging Face or accessed via major platform partners, including AMD, AWS, Databricks, Dell, Google Cloud, Groq, IBM, Intel, Microsoft Azure, NVIDIA, Oracle Cloud, Snowflake, and more.
Image Editing— the new Imagine edit feature using meta.ai allows you to edit photos using natural language, such as adding new backgrounds and changing colors.
Image Reasoning - The Llama 3.2 1B and 90B models support image reasoning use cases, such as document-level understanding, charts and graphs, image captioning, and visual grounding tasks, such as directionally pinpointing objects in images based on natural language descriptions.
On-Device Use Cases - The Llama 3.2 1B and 3B models support context length of 128K tokens and are state-of-the-art in their class for on-device use cases like summarization, instruction following, and rewriting tasks running locally at the edge.
Private, High-Performance AI - Running the 1B and 3B models locally enables two major advantages. First, prompts and responses can feel instantaneous since processing is performed locally. Second, running models locally is more secure and enforces privacy by not sending data to the cloud. Since processing is handled locally, the application can control which queries stay on the device and which may need to be processed by a larger model in the cloud.
The Value Stick - How to Create Business Value Using AI
What
The objective of data and AI strategy is to derive value; it’s not a science project. It’s common for people to say that data and AI will create value, but they are unclear exactly how that will happen. They will broadly state that it will create a “competitive advantage" but rarely in context to Michael Porter’s definition:
Competitive advantage allows you to follow the precise link between the: value you create, how you create it (your value chain), and how you perform (your P&L). If you have a real competitive advantage, it means compared with rivals, you: operate at a lower cost, command a premium price, or both.
Deciding where to focus your data and AI strategy can be challenging. A value-based framework called the “value stick” can help you determine the optimal use cases for deriving value.
The “value stick” is a strategic decision framework that states an initiative must pass one of these tests, or it’s not worth doing:
Creates value for customers by raising their willingness to pay (WTP).
Creates value for employees by making work more appealing.
Creates value for suppliers by reducing their operating cost.
These tests are captured in a value stick - a strategic framework introduced by Felix Oberholzer-Gee in his book Better Simpler Strategy. It is a visual and conceptual tool that helps businesses understand how value is created and divided among the company, its customers, and its suppliers.
The value stick simplifies complex economic interactions into four key components:
Willingness to Pay (WTP): The maximum price a customer is willing to pay for a product or service.
Price: The actual amount the customer pays.
Cost: The company’s expenditure to produce the product or service.
Willingness to Sell (WTS): The minimum price at which a supplier is willing to provide inputs.
By plotting these components on a vertical line—the “stick”—businesses can visualize the distribution of value:
Customer’s Share (Consumer Surplus): The difference between the customer’s willingness to pay and the actual price paid (WTP - Price).
Company’s Share (Profit Margin): The difference between the price received and the cost of production (Price - Cost).
Supplier’s Share (Supplier Surplus): The difference between the cost of inputs and the supplier’s minimum acceptable price (Cost - WTS).
Strategy - How to Use It
Here are a few use cases where you can use AI to effectively manipulate the value stick components to create and capture more value.
↗️ Increasing Willingness to Pay (WTP)
Personalized Customer Experiences:
AI-Driven Recommendations: Utilizing machine learning algorithms to analyze customer data and provide personalized product or content recommendations. Example: E-commerce platforms like Amazon use AI to suggest products customers are more likely to purchase, enhancing the shopping experience.
Chatbots and Virtual Assistants: Implementing AI-powered chatbots to provide instant customer support and improve service quality.
Enhanced Product Features:
Smart Product Features: Integrating AI into products to add intelligent features that enhance functionality. Smart thermostats like Nest learn user preferences to optimize home heating and cooling, increasing WTP.
Improved Customer Insights:
Predictive Analytics: Using AI to predict customer behavior and tailor offerings accordingly. Example: Netflix uses AI to predict a user's favorite shows or movies, increasing engagement and WTP.
↘️ Decreasing Willingness to Sell (WTS)
Optimized Supply Chain Management:
AI-Powered Forecasting: Enhancing demand forecasting to reduce inventory costs and improve supplier negotiations. Example: Retailers use AI to predict demand spikes, allowing suppliers to plan production more efficiently.
Supplier Relationship Management:
Automated Procurement Processes: Streamlining procurement with AI to reduce administrative burdens on suppliers. Example: AI systems can automate order placements when inventory is low, providing suppliers with consistent business and reducing their WTS.
Dynamic Pricing for Suppliers:
Transparent Pricing Models: Using AI to analyze market conditions and offer suppliers fair pricing, encouraging them to lower their WTS. Example: Ride-sharing platforms adjust driver payouts based on demand, incentivizing drivers to offer services during peak times.
↘️ Reducing Costs
Operational Efficiency:
Process Automation: Implementing AI-driven process automation for repetitive tasks to reduce labor costs and errors. Example: Manufacturing companies use robotic process automation (RPA) to handle assembly line tasks.
Energy Management: Using AI to optimize energy consumption in facilities. Example: Google uses DeepMind AI to reduce data center cooling costs by up to 40%.
Improved Decision-Making:
Data Analytics: Leveraging AI to analyze large datasets for better strategic decisions, reducing costly mistakes. Example: Financial institutions use AI to detect fraudulent transactions, saving costs associated with fraud.
↕️ Optimizing Price
Dynamic Pricing Strategies:
Real-Time Price Adjustments: Using AI algorithms to adjust prices based on demand, competition, and other market factors. Example: Airlines and hotels adjust pricing in real-time to maximize revenue based on booking patterns.
Enhanced Value Proposition:
Bundling and Custom Offers: AI helps create personalized bundles or offers that align with customer preferences, allowing optimal pricing. Example: Streaming services suggest subscription bundles based on user viewing habits.
The convergence of voice AI and large language models like Llama 3.2 presents exciting possibilities for natural interaction. ChatGPT Advanced Voice leverages these models to create more human-like conversational experiences. Applying the Value Stick framework can help businesses identify specific use cases where AI can deliver tangible benefits, such as improved customer service or personalized marketing. You talked about ChatGPT Advanced Voice in your post. How do you see the integration of acoustic modeling techniques within Llama 3.2 impacting the expressiveness and naturalness of generated speech? Imagine a scenario where a user needs to control a complex scientific instrument through voice commands, relying on AI to interpret nuanced instructions and execute precise actions. How would you technically use ChatGPT Advanced Voice and Llama 3.2 to build a system capable of handling such intricate interactions?
Analytics Leader | Analytics & Data Strategy | Data Science | Certified Chief Data & AI Officer - Carnegie Mellon
7moI'm especially enjoying the convergence of traditional human centered data driven design, augmented through AI. We are truly in amazing times.
Data Product Leader | Driving Business Impact with Data, AI, ML, Analytics & Platform Strategy | Built & Led Full-Stack Data Teams | Speaker | Advisor
7moVery informative read James. Willingness to pay is a great strategy that solves for customer acquisition and retention.
Co-Founder of Altrosyn and DIrector at CDTECH | Inventor | Manufacturer
7moThe convergence of voice AI and large language models like Llama 3.2 presents exciting possibilities for natural interaction. ChatGPT Advanced Voice leverages these models to create more human-like conversational experiences. Applying the Value Stick framework can help businesses identify specific use cases where AI can deliver tangible benefits, such as improved customer service or personalized marketing. You talked about ChatGPT Advanced Voice in your post. How do you see the integration of acoustic modeling techniques within Llama 3.2 impacting the expressiveness and naturalness of generated speech? Imagine a scenario where a user needs to control a complex scientific instrument through voice commands, relying on AI to interpret nuanced instructions and execute precise actions. How would you technically use ChatGPT Advanced Voice and Llama 3.2 to build a system capable of handling such intricate interactions?