This is a continuation of my previous post. There is more than I thought. I thought I could finish it all at once by just listing what I organized... I'm already afraid that it will be seen as a boring study.
In fact, there are various ways to use artificial intelligence. Currently, use cases and services are pouring out mainly for generative AI, so the way of using it is not much different. Let's take a look at some representative services.
Summary: How to use AI to summarize content
The summary pattern is a feature that is being used more effectively due to the advancement of generative AI. This pattern helps you quickly summarize long documents or conversations to focus on important information. It is useful for quickly grasping the key points in everyday tasks such as organizing meetings, reading documents, and preparing for conversations.
This method is often used for summarizing tasks, email summaries, meeting recording summaries, etc.
The summary pattern is used in various ways, such as extracting key content from long documents, synthesizing multiple emails, and saving automated summaries. The main advantages of this method are time savings, automated workflow, and the possibility of deeper exploration. In fact, Liner in Korea is a good example of applying AI appropriately to pages that you usually underline or save. However, there is a risk of missing important information due to excessive simplification, so it is important to provide the original information source. In fact, it is the most used and commercialized function. Because of the nature of LLM, it is the easiest to implement and difficult to disappoint customers...
Blend: How to combine prompts and new elements to drive creative results
Remixing and Blending patterns are ways of using AI to combine multiple prompts or sources to create new results. This method allows users to fine-tune AI results and achieve unexpected creative results.
The main advantage of this pattern is the creative use cases and the learning and creativity through play. Users can combine different sources to create unique results, and the prompts can be gradually developed to increase engagement and interest. However, adding too much information or using complex prompts can lead to confusing results, so it is important to help users improve their prompt writing skills.
Starting with Midjourney , it is called by various names such as Remix and Blend, but in the end, it mixes or modifies existing data to create something completely new.
Auto Fill: How to easily expand multiple inputs with one prompt
Auto Fill is an AI feature that automatically fills in multiple data items with a single prompt. Sometimes it is implemented to work with just Context without prompts. Auto Fill is particularly effective in database or spreadsheet work. It is a feature that shows the agentic nature of AI well, allowing users to focus on more creative and strategic work by automating repetitive tasks.
The desire for automatic filling actually goes back to the history of OA, with Excel, mail merge, etc. Recently, Adobe has shown that filling is not limited to simply working with tables, with inpainting and outpainting being freely available, making it a frequently used method.
Inline Action: How to use it contextually with content available on the page
Inline Actions are a feature that allows AI to interact with existing content on the page. Users can select specific text to edit or add new elements, allowing for natural interaction with AI and fine-grained control. The main advantage is greater user control and context-based interaction. Users can focus on only what they need and adjust AI output, and fine-tune results to fit specific contexts.
However, there is also a potential risk that AI products have difficulty tracking connections to previous work. Inline Actions are particularly well-suited for text editing or references, where AI can modify and supplement existing work. This pattern allows AI to be used more flexibly and contextually, resulting in a more intuitive and precise user experience. I have yet to find a better use case for Notion AI . The idea of suggesting things naturally while writing is secretly difficult to implement. I thought of it while typing, and Copliot, which is used in VScode, is also good because it can be used naturally. Feels like auto-completion?
Synthesis: A method of reorganizing complex information into a concise structure for use.
I said Synthesis, but honestly, it's a fancy way of saying Paraphrasing. It's an AI function that reconstructs complex information from multiple sources into a concise and meaningful structure. It goes beyond simple summarization and systematically reorganizes data so that it can be presented to users as new content while maintaining the essence.
The main advantages are ease of use and multimodality. It can be summarized and reorganized with simple operations without complex prompts, and the results can be provided in various formats. However, it is difficult to clearly convey the limitations of AI to the user, so there is a risk of making incorrect conclusions. However, it is good to use when organizing or formatting raw data.
Share your progress
The case of Naver Cue is reproduced when asking questions related to shopping or travel.
This is a way to transparently show the AI's thinking process to the user. This solves the 'black box' problem of AI work and enables user understanding and intervention. It mainly appears in two forms, 'Show my work' and 'Check my work', the former showing the AI's response generation process, and the latter showing the area that AI will affect in advance.
The main advantage of this pattern is the transparency before the action, which allows the user to gain trust and maintain control over the AI. It also provides the user with an opportunity to learn how the AI is approaching things. However, there is also a potential risk that it can be an unnecessary waste of time. However, it is a meaningful approach in recent times in terms of increasing trust and showing the process.
Step 4: Feedback (User Feedback)
It is used and evaluated. Or it is verified. It checks whether the command I gave was properly carried out and whether the so-called hallucination occurred. Recently released artificial intelligence services provide results in real time, so people can provide feedback right away, and it is progressing more quickly.
You could call it RLHF (Reinforcement learning from human feedback), but in reality, not many people are doing it. It's harder than you think. (To be precise, human feedback is easy, but reinforcement learning is difficult.)
Source function
The citation function plays a key role in this process. It allows AI to trace the source of information provided and helps users transparently check the logic and basis of AI. For example, Adobe PDF Summarize highlights specific passages in a document, and Perplexity AI and Bing Copilot provide summary information referencing multiple external sources.
These features are implemented differently on each platform, but they commonly contribute to improving the user's understanding of information. As a result, the interaction between AI and users is becoming richer and more reliable, and the importance of these features is expected to increase further as AI technology advances in the future. It is also the fastest way to overcome hallucination.
Creation control function
As AI grows in complexity, it becomes essential to provide users with control over the AI’s information flow, allowing them to stop or reset requests when necessary.
Controls like the stop icon allow users to take an active role in their interactions with AI. This makes AI an interactive partner rather than a mere tool, improving the user experience and increasing system efficiency.
Positive/negative display function
The evaluation system of AI models is an important tool for improving models through user feedback. It is usually implemented as a thumbs up or star rating, and has the advantage of providing real-time feedback and enhancing user authority. If the evaluation does not immediately lead to improved experience, users may regard it as simple data collection.
Therefore, the evaluation system must be transparent and provide real improvements to users, which will enhance the interaction between users and models and lead to continuous improvement of the experience.
Regeneration
The AI response regeneration feature greatly improves the user experience by generating multiple responses to the same prompt. Key benefits of this feature include providing multiple options, more control for the user, the opportunity to explore new ideas, and the ability to track and compare previous results.
For effective implementation, it is important to preserve the first response, provide learning opportunities through understanding AI logic, apply to various situations, and maintain consistency through user-controlled parameters. This regenerative function makes the interaction with AI more flexible, helping users achieve optimal results.
Provide response options
The ability to provide response choices is an important tool for enhancing and increasing the efficiency of AI-user interactions. This feature gives users the power to choose between multiple AI responses. Key benefits include facilitating a user-driven learning process, increasing user control, and improving AI adaptability.
By selecting the optimal result from a variety of responses, users can increase AI’s understanding and reach the desired result faster. AI also better understands the user’s preferences and intentions. However, it is necessary to consider managing the user’s frustration when failure occurs, the need for tracking the selection process, and maintaining consistency in the user experience. In conclusion, this feature provides users with a powerful tool to compare and select AI responses, accelerating AI learning and achieving more accurate and personalized results. This allows AI to build greater trust in its interactions with users.
Due to a failure in portion control, it will be continued in Part 3.
📚 Welcome to Haebom's archives. --- I post articles related to IT 💻, economy 💰, and humanities 🎭. If you are curious about my thoughts, perspectives or interests, please subscribe. Would you like to be notified when new articles are posted? 🔔 Yes, that means subscribe. haebom@kakao.com
