Edges of Innovation - #3

Welcome to the third issue of "Edges of Innovation", where we delve into the dynamic realms of innovation and technological advancements. In this issue, "Leveraging Existing Networks for Breakthrough Innovation ??" unravels the often-overlooked strategy of utilizing existing frameworks for groundbreaking innovations. Far from the common belief that innovation requires creating something entirely new, we showcase real-world examples where companies have achieved significant success by creatively repurposing and aligning with pre-existing systems.

In the second feature, "Transform Your Work: OpenAI Launches Tailor-Made GPTs", we introduce OpenAI's latest innovation - custom ChatGPT models, or GPTs. These personalized AI models mark a significant step towards making advanced AI accessible and useful across diverse fields. Whether enhancing educational tools, streamlining business workflows, or revolutionizing customer support, these tailor-made GPTs open up endless possibilities for customization and efficiency. Plus, the upcoming opportunity for developers to monetize their GPTs in the new App store adds an exciting dimension to this technological leap. Prepare to dive into a world where AI is not just a tool but a customizable partner in your personal and professional growth. Stay tuned for insights that could redefine how you interact with AI and its capabilities in everyday life!

Leveraging Existing Networks for Breakthrough Innovation ??

A recent article from the Harvard Business Review, The “Piggyback” Approach to Innovation , created the perfect opportunity to come back on the notion of innovation. The Merriam-Webster dictionary tells us that innovation is the introduction of something new, while Wikipedia tells us that innovation?is the practical implementation of?ideas?that result in the introduction of new goods or services or improvement in offering goods or services.

In a world where innovation is often synonymous with starting from scratch, the "Piggyback Approach", described in the HBR article, offers a refreshing perspective. The article reminds us how businesses are harnessing existing systems to drive growth and impact:

?

?? ColaLife saw Coca-Cola's distribution prowess and asked, "Why not medicines?" They designed kits to fit between soda bottles, bringing life-saving treatments to remote areas.

?? Wise (formerly TransferWise) tapped into the global banking system, offering fair currency exchange and low fees, turning a personal pain point into an $11 billion valuation.

?? Safaricom transformed Kenya's mobile network into a banking revolution with M-Pesa, proving that financial inclusion can be a text message away.

?? Nestlé fortified common foods with essential nutrients, addressing micronutrient deficiencies while boosting their bottom line.

?? Paramount Pictures cleverly piggybacked on the "50 Shades of Grey" hype with a SpongeBob twist, showcasing the power of strategic marketing alignment.

When you think about the Facebook, AirBnB or Google of the world, they never "invented" something new. They started from something that existed and twisted it. The takeaway? Innovation doesn't (generally) mean reinventing the wheel (it rarely does). Often, it's about finding a new way to ride the ones already turning.

?

?? For entre/intrapreneurs and leaders, here's the action plan:

1. Map existing networks and systems relevant to your industry.

2. Identify underutilized assets or capabilities within these systems.

3. Develop creative solutions that align with these pre-existing networks.

4. Test, learn, and iterate rapidly (using an MVP approach, as we will see in a future post).

?

Innovation is everywhere – sometimes, it just takes a piggyback ride to see it.

?#InnovationStrategy #BusinessGrowth #PiggybackInnovation #Entrepreneurship

Transform Your Work: OpenAI Launches Tailor-Made GPTs!

OpenAI's latest announcement introduces "GPTs" – a revolutionary step that allows individuals and businesses to create custom versions of ChatGPT for specific purposes. These tailored AI models can be designed to assist with a wide range of tasks, enhancing efficiency and personalization.

What are GPTs?

GPTs are specialized instances of ChatGPT that you can train for particular tasks or industries. They can be as niche or as broad as needed, from helping you learn the rules of a new board game to teaching math or designing creative content. The beauty of GPTs lies in their simplicity: no coding is required, making AI customization accessible to everyone.

There are thousands potential use cases in every industry. For instance:

1. Educational Tools: Craft a GPT to serve as a personal tutor, providing customized lessons and adapting to different learning styles.

2. Professional Productivity: Implement a GPT to manage business workflows, automate repetitive tasks, and streamline communication.

3. Enhanced Customer Support: Deploy a GPT as a customer service agent, trained to understand your products deeply and provide instant, accurate responses.

4. Creative Collaborator: Integrate a GPT with design software to offer real-time suggestions, generate ideas, or even create initial drafts for creative projects.

5. Data Analyst: Utilize a GPT to sift through data, identify trends, and offer actionable insights, effectively acting as an on-demand data scientist.

Actionable Insights:

- Start by identifying areas in your life or business where efficiency could be improved through automation.

- Visit chat.openai.com/create to explore the process of creating your own GPT.

- Consider the example GPTs, such as those integrated with Canva for design or Zapier for workflow automation, to spark ideas.

- Keep being curious... the future is exciting!

By harnessing the power of GPTs, you can unlock new levels of productivity and innovation. Whether for personal use or enterprise solutions, the potential is vast and waiting to be explored.

One last thing... Developers will be able to monetize their GPTs in the forthcoming App store. This opens a new way to build specialized conversational AIs with deep expertise. Of course, this does not come with a set of challenges and questions regarding the corpuses that will be used to build those specialized AI, but it's really exciting for experts of a domain to create a kind of a doppelganger, while monetizing it.

How will you use GPTs? Let us know in the comments.

#OpenAI #GPTs #AI #Customization #Innovation #Technology #BusinessEfficiency #AIForEveryone

Limits of ChatGPT? RTFM!

?? Some people (including self-called experts) get so excited by the capabilities of ChatGPT, they sometimes tend to exaggerate what this conversational AI is capable of. This is particularly true for those who are not developer and have no technical knowledge on how an AI model is created. When you're a developer, you know that the truth always lies in the documentation. So, when in doubt, developers RTFM (Read The Fine Manual, this is the politically correct definition of RTFM, the real one is using the F word ??).

True Limits

After OpenAI DevDay , I RTFM and consulted the updated OpenAI documentation, and specifically the part about the models . Hence, here's a clear-cut guide to the limits of OpenAI's GPT-3.5, GPT-4, and the newly announced GPT-4 Turbo:

?? GPT-3.5

- Context window: Up to 16,385 tokens

- Output window: Maximum of 4,096 tokens

- Training data cut-off: Up to Sep 2021

- Best for: Natural language and code generation

- Highlight: GPT-3.5 Turbo is fine-tuned for chat applications with enhanced instruction following. It's the model available free of charge through https://chat.openai.com .

?? GPT-4

- Context window: 8,192 tokens (standard), 32,768 tokens (GPT-4-32k)

- Output window: Maximum of 4,096 tokens

- Training data cut-off: Up to Sep 2021

- Best for: Advanced reasoning and broader knowledge tasks

- Note: GPT-4 outperforms GPT-3.5 in complex reasoning scenarios. It's available to Plus and Enterprise users (paid).

?? GPT-4 Turbo

- Context window: 128,000 tokens

- Output window: Maximum of 4,096 tokens

- Training data cut-off: Up to Apr 2023

- Best for: High-level reasoning with natural language or code, and image inputs (GPT-4 Vision)

- Caution: Currently a preview model and not yet suited for production environments. It's only available (for the moment) through the API for paying developers.

Tokens, Window, Training and What to Remember

Ok. You see a few things in the text above that call for a clear and concise information:

?? Token: "In English, a token can be as short as one character or as long as one word (e.g., a or apple)". This is from OpenAI documentation (RTFM). But it's confusing as we don't know how many tokens a text is made of. Well, we can take a fair guess and say a token is a 3/4 of a word or a punctuation sign. (If you don't want to guess, OpenAI advise you to use the tiktoken function).

?? Window: the input and the output parts of a conversation. Therefore, when we say the context window is 16,385 tokens we can assess that the addition of our question + ChatGPT's answer cannot be longer than 16,385 tokens. So, if you ask a 16,000-token long question, the answer cannot be longer than 385-token long. In all cases, the length of an answer (output window) cannot be longer than 4,096 tokens (less than 4k words).

?? Training: ChatGPT is an artificial neural network model, based on a Generative Pre-trained Transformer (GPT, yes, this is what it stands for) architecture. Calculating an artificial neural network model requires a lot of computation based on existing preformatted data (great explanation here if you have some time). Once the model is calculated (that means the weight on each artificial neuron is locked), it can be used, but cannot, generally, be updated easily. This explains why ChatGPT 3.5 and 4 training data cut-off is September 2021, and still is, though the new ChatGPT 4-Turbo model has been calculated with a new set of data with an April 2023 cut-off date. So no, all models are equal when it comes to cut-off.

??? One last point to remember. There's a difference between your ChatGPT client (the one you use when you go to https://chat.openai.com ) and the model that is available directly through the API (requiring coding). Hence, as I write this post, Plus and Entreprise users have access through the chat client to GPT-4 and NOT GPT-4 Turbo ("GPT-4 Turbo is available for all paying developers to try by passing gpt-4-1106-preview in the API and we plan to release the stable production-ready model in the coming weeks.", source: https://openai.com/blog/new-models-and-developer-products-announced-at-devday ), as the latest is available through the API only for the moment.

I hope this clarifies all so you can run the right model in the chat window, or build you own chat window/program through the API.

My last advice: RTFM! ?? Never believe anybody who provide statements without citing his/her source. ??

?? Selecting the right model is pivotal for achieving your goals. Whether it's engaging in natural conversations, tackling intricate problems, or processing extensive contexts, there's a GPT variant ready for the task.

Stay informed and choose wisely to power your projects. ??

#AI #OpenAI #GPT4 #MachineLearning #Technology

Revolutionize Chatbots with OpenAI's AI Assistants

?? Elevate Your Business with Next-Gen AI Assistants

OpenAI announced the new Assistants API, a game-changer for businesses looking to integrate advanced chatbots. Here's why.

What are AI Assistants?

AI Assistants are sophisticated chatbots that can perform a wide range of tasks, from solving complex math problems to creating data visualizations. They're powered by the latest in AI technology, including GPT-4, and can be customized with specific instructions to fit your business needs.

How Can They Help Your Company?

- Personalization: Tailor the assistant's personality and capabilities to your company's voice and objectives.

- Multitasking: Access multiple tools in parallel, including code interpretation and knowledge retrieval.

- Persistence: Maintain context over long conversations with smart message truncation.

- Rich Content: Work with various file formats and generate rich content like images and spreadsheets.

Benefits Over Previous Generations

- Advanced Contextual Understanding: Handle complex queries with nuanced understanding.

- Greater Flexibility: Customize with up to 128 tools for a wide array of tasks.

- Efficient Conversation Management: Smart truncation ensures relevant context is always in focus.

- Dynamic Content Creation: Generate and reference files within conversations for richer interactions.

Use Cases

1. Automated Technical Support: AI Assistants can troubleshoot technical issues by guiding users through diagnostic steps or directly interfacing with system APIs to perform checks and suggest solutions.
2. Sales and Lead Qualification: Use AI to engage potential leads, assess their interest level, and gather preliminary information to streamline the sales process.
3. Personalized Recommendations: Implement AI Assistants that analyze user preferences and past behavior to suggest personalized products, services, or content.
4. Language Tutoring: Create language learning assistants that provide personalized grammar exercises, vocabulary training, and conversational practice.
5. Healthcare Assistance: Develop AI Assistants that help patients manage their health by reminding them to take medications, scheduling appointments, and providing basic health advice.
6. Financial Advising: Assistants can analyze financial data to provide personalized investment advice or budget planning.
7. Event Planning and Coordination: AI Assistants can help organize events by managing RSVPs, sending reminders, and providing attendees with information.
8. HR and Recruitment: Automate initial candidate screening, answer FAQs about positions, and schedule interviews.
9. Travel Concierge Services: Offer AI-powered travel assistance, providing itinerary planning, booking services, and real-time travel updates.
10. Interactive Storytelling: Engage users with interactive stories or games where they can choose their own adventure, guided by the AI.
11. Customer Support: Provide detailed, context-aware solutions to customer inquiries.
12. Education: Tutor students in various subjects with interactive problem-solving.

Key Actions to Take

1. Explore the API: Familiarize yourself with the Assistants API capabilities.

2. Identify Use Cases: Determine how AI Assistants can best serve your business needs (the above examples can help).

3. Integrate and Customize: Build your assistant with your preferred development tool.

4. Iterate and Improve: Use feedback to refine the assistant's performance.

Ready to harness the power of AI Assistants for your business? Dive into the Assistants API and start building your conversational AI chatbot today!

#AI #Chatbots #BusinessGrowth #Innovation #OpenAI

Crazy, Exciting and Latest News

• Humane AI Pin has been announced as the end of the smart phone as we know it. https://www.youtube.com/watch?v=XdAlAI9TLww and https://vimeo.com/882968794 . It's true that it's pretty amazing and leverages the power of Generative AI. Let's see how this unfolds.
• If you happen to be in Paris on Nov 17, AI Pulse is happening at Station F. https://www.ai-pulse.eu . A day not to miss.
• Open Source models are blossoming with the new addition of Yi, from 01.ai , the company of Kai-Fu Lee, Author of 'AI Superpowers' and 'AI 2041'. More on Open Source AI models in a forthcoming issue of Edges of Innovation.
• Microsoft Ignite is happening in Seattle this week also. https://ignite.microsoft.com/en-US/home . Watchable online, I expect a lot of new things around AI and Azure OpenAI stack.
• Cryptocurrencies are back up, with ETH going above 2k USD. Good news for hodlers and all blockchain aficionados.

Voilà, this is all for this week. Stay tuned for next, innovation and AI are hot topics these days...