Things I know about GPT Models MS & OpenAI
What can ChatGPT does that?
Summarize articles
Generate creative writing, including poetry and stories
Answer questions based on the knowledge it was trained on
Simulate conversation in a chatbot application
ChatGPT’s weakness
Training data is only until 2021. This means that it does not have access to real-time information or events that happened after this date.
However, system design can help mitigate this. For example, when integrated with a browser tool, ChatGPT can search the internet to find more recent information. But please note that this isn't a native functionality of GPT itself.
Who made ChatGPT and how is it related to Microsoft?
ChatGPT was created by OpenAI.
Microsoft has invested in OpenAI and also provided them with computing resources.
Microsoft has an exclusive license to use the GPT-3 model.
Kinds of AI made by OpenAI
GPT-3 and GPT-4: Models that generate and understand text.
Microsoft provides these models on Azure.
Codex: A model that understands and generates code.
Microsoft provides this model on Azure as well.
DALL·E: A model that generates images from text prompts.
Using OpenAI’s models on Azure
Azure OpenAI Studio: This service supports the use of OpenAI models. Developers can execute these models in a "PlayGround" environment.
Use cases for ChatGPT
Analyzing customer service interactions.
Performing SWOT analysis within Excel by using the ChatGPT API.
Difference between Fine-Tuning and Embeddings
Fine-Tuning: Adjusting the parameters of the GPT models to better fit a specific task or dataset.
Embeddings: Representations of words or phrases within the model's vector space that capture semantic meaning.
Prompt Engineering
Structure of a Prompt: A prompt typically includes:
Context: Information the model needs to understand the prompt.
Instructions: What the model is being asked to do.
Exemplar: A model of what a correct response looks like.
Input Data: The primary content for the model to process.
Output Indicator: Signal for the model to generate a response.
Difference in responses based on parameters
Symptoms Urgency Confidence Prompt English Translation Hand trembling, legs shaking 0.7 0.95 I held their hand and said, "Don't let go." I felt that if I let go, they would fall, so I desperately held on. That's great. Your actions are admirable. Supporting someone who is about to fall is a necessary action to provide help. Please continue to act in the same way and provide assistance safely. Hand trembling, legs shaking 1 0.6 I held their hand and said, "Don't let go." I felt that if I let go, they would fall, so I desperately held on. Your attitude is excellent. You used all your strength to support the other person and overcome the situation. You have confidence in your abilities and can calmly handle difficult situations. You are very reliable. Please continue to maintain that attitude and support the people around you. Hand trembling, legs shaking 0 0.95 I held their hand and said, "Don't let go." I felt that if I let go, they would fall, so I desperately held on. Your attitude is very admirable. It's wonderful to have the courage and decisiveness to put yourself in danger to protect others. However, it's important to act safely and avoid such situations. Also, don't forget to seek help from others when needed.
About Copilot
Microsoft 365 Copilot: This assistant is built on OpenAI's GPT-4 large language models and incorporates Microsoft Graph. It translates user text input into content in Microsoft 365 apps like Word, Excel, PowerPoint, Outlook, and Teams. Here's what it can do in different applications:
In Word, it can generate and edit text based on user prompts, and even suggest rewrites to strengthen arguments**1**.
In Excel, it can assist with data analysis by formatting data, creating graphs, generating PivotTables, identifying trends, and summarizing information**1**.
For PowerPoint, it can create presentations based on information from selected Word documents and Excel spreadsheets or user prompts**1**.
In Outlook, it can draft emails with varying length and tone based on user input and pull relevant information from other emails**1**.
In Teams, it can present information for upcoming meetings, transcribe meetings, and provide debriefs for users who join the meeting late**1**.
Other tools like OneNote and Viva Learning also use Copilot for drafting plans, generating ideas, and creating personalized learning journeys**1**.
Microsoft Dynamics 365 Copilot: This assistant provides interactive, AI-powered assistance across business functions. For instance, it can help sellers reduce time spent on clerical tasks, empower agents to deliver exceptional customer care, and simplify marketers' workflows in data exploration, audience segmentation, and content creation. It also streamlines the creation of product listings for online commerce in Business Central, and in the Supply Chain Center, it can proactively flag external issues that may impact key supply chain processes**2**.
MS Power App with Copilot
MS Security Copilot
MS GitHub copilot: GitHub Copilot is an AI-powered code completion tool developed by GitHub in collaboration with OpenAI. It is designed to work as an AI pair programmer that helps you write code faster and with fewer errors. It provides suggestions for whole lines or blocks of code as you type, and works with a broad set of frameworks and languages, although its capabilities are strongest in Python, JavaScript, TypeScript, Ruby, and Go. Copilot's suggestions are based on patterns found in publicly available code, so it doesn't know about the specifics of your private codebases unless you have been typing similar code during your current coding session. It is important to note that while GitHub Copilot can generate useful and time-saving code, it doesn't guarantee the correctness of the suggestions and developers should always review the code suggestions before using them1.
この記事が気に入ったらサポートをしてみませんか？