Sign In

AI UX Study Sharing Session 2/3

Haebom
Step 3: Using (Using Products/Services)Summary: How to use AI to summarize contentBlend: How to use prompts and new elements together to drive creative resultsAuto Fill: How to easily expand multiple entries with a single promptInline Action: How to use it contextually with content on the pageSynthesis: How to reorganize complex information into a simplified structure for useSharing the processStep 4: Feedback (User Feedback)Source attribution featureGeneration control featureThumbs up/down indicatorRegenerateOption to choose responses
Continuing from the previous post. There’s a lot more than I expected. I thought it would be a one-shot deal if I just listed what I organized... I’m already worried it might come across as a boring study session.

Step 3: Using (Using Products/Services)

There are actually a variety of ways to use artificial intelligence. Right now, most cases and services are centered on generative AI, so the methods of use aren’t that different. Let’s take a look at some representative services.

Summary: How to use AI to summarize content

Summary patterns are being used even more effectively thanks to advances in generative AI. This approach helps summarize long documents or conversations so you can focus on the important information. It’s handy for quickly grasping the main points during everyday tasks like meeting recaps, reading documents, or prepping for conversations.
이런 식으로 Task를 요약하는 방법 부터, 이메일 요약, 회의 녹음 요약 등으로 자주 쓰입니다.
Summary patterns are utilized in a variety of ways, including extracting key content from long documents, synthesizing multiple emails, and automatically saving summaries. The main advantages of this method are time savings, automated workflows, and the potential for deeper exploration. In fact , Liner in South Korea is a good example of applying AI appropriately to pages that users have previously underlined or saved. However, oversimplification can lead to the omission of important information, so providing the original source of the information is crucial. In fact, this is the most widely used and commercialized feature. Given the nature of LLM, it's also the easiest to implement and the least likely to disappoint customers...

Blend: How to use prompts and new elements together to drive creative results

Remixing and Blending patterns are ways to leverage AI by combining multiple prompts or sources to generate new results. This lets users fine-tune what AI creates and even get unexpected, creative outcomes.
This approach’s main strengths are its creative applications and the chance to learn and be imaginative through play. By blending diverse sources, users can create unique outcomes, and by gradually improving prompts, they can boost engagement and interest. However, adding too much info or overly complex prompts can sometimes lead to confusion, so helping users sharpen their prompt-writing skills is key.
Starting with Midjourney, it is called by various names such as Remix and Blend, but in the end, it is about mixing or modifying existing data to create something completely new.

Auto Fill: How to easily expand multiple entries with a single prompt

Auto Fill is an AI feature that automatically populates multiple data fields with a single prompt. In some cases, it’s even set up to work just with context, no prompt needed. Auto Fill makes database and spreadsheet tasks much more efficient. By automating repetitive jobs, it really showcases AI’s agent-like qualities, letting users focus on more creative and strategic work.
The desire for auto-filling has been around as long as the history of office automation—think Excel or mail merge. Lately, Adobe has made features like inpainting and outpainting readily available, showing that filling things in isn’t just limited to spreadsheets anymore; it’s become a frequently used method all around.

Inline Action: How to use it contextually with content on the page

Inline Action is a feature that lets you interact with AI based on existing content on the page. You can pick specific text to edit or add something new, allowing natural AI interactions and fine-tuned control. The key benefits here are enhanced user control and context-aware interactions. Users can focus only on the parts they need and tweak AI outputs precisely for certain contexts.
However, there's also a potential risk that AI products struggle to track connections with previous work. Inline Actions are particularly well-suited for text editing, referencing, and for AI to modify and supplement existing work. This pattern allows for more flexible and contextual use of AI, resulting in a more intuitive and precise user experience. I haven't yet found a better use case for Notion AI . Naturally suggesting actions while writing is surprisingly difficult to implement. It occurred to me while typing, and Copliot, used in VS Code, is also great because it's so easy to use. It feels like auto-completion, right?

Synthesis: How to reorganize complex information into a simplified structure for use

We call it Synthesis, but honestly, it sounds like a fancier way to say Paraphrasing. This AI function takes complicated information from multiple sources and turns it into something more concise and meaningful. It’s more than just summarizing—it systematically reorganizes data, so you get what feels like new content, even though the core stays the same.
Its main advantages are ease of use and being able to work with multiple formats. You can summarize and reorganize things with just a few steps—no need for complicated prompts—and get results in all sorts of forms. But it is hard to clearly show users where AI’s limitations are, so there’s a risk of bad conclusions. Still, it’s great when you need to structure or format raw data.

Sharing the process

네이버 Cue의 사례, 쇼핑이나 여행 관련 질문을 할 경우 재현 됩니다.
This is a way to make the AI’s thought process transparent to the user. It tackles the ‘black box’ problem in AI, allowing people to understand and get involved. You’ll mostly see two forms: ‘Show my work’—which reveals how the AI’s response is put together—and ‘Check my work’—which previews the areas the AI is about to affect.
The main advantage of this pattern is transparency before anything happens, allowing users to trust the AI and keep control. It also gives people a chance to understand how the AI works. On the flip side, it could waste time if it gets too detailed. Still, these days it’s a meaningful way to build trust and show what’s happening under the hood.

Step 4: Feedback (User Feedback)

Once you use it, you get to evaluate or review it. You check whether the instructions were properly carried out, or if any hallucinations happened, as people often say. Since most of the newest AI services provide real-time results, people can give feedback instantly, and this is driving even faster progress.
You could fancy it up by calling it RLHF (Reinforcement Learning from Human Feedback), but the truth is, not many places actually do it this way. It’s trickier than you’d expect. (To be exact, collecting human feedback is easy; the reinforcement learning part is what’s tough.)

Source attribution feature

The citation feature plays a key role in this process. It allows users to trace the source of information provided by AI and provides transparency into the AI's logic and rationale. For example, Adobe's PDF Summarization feature highlights specific passages in a document, while Perplexity AI and Bing Copilot provide summaries that reference multiple external sources.
These features are implemented differently across platforms, but in every case they help users better understand information. As a result, interactions between AI and users are becoming richer and more trustworthy, and as AI tech keeps advancing, these features will only become more important. It’s also the quickest way to address hallucinations.

Generation control feature

As AI becomes more complex, giving users control features has become essential. These let users manage how the AI processes info, and stop or reset their requests if needed.
Controls like a stop icon let users take a proactive role when interacting with AI. These features help people see AI as an interactive partner, not just a tool—making for a better user experience and improved system efficiency.

Thumbs up/down indicator

The evaluation system for AI models is a key tool for improvement through user feedback. It’s usually built as a thumbs-up/down or star rating, giving the benefits of real-time feedback and boosting user agency. If users don’t see the benefits right away, though, they might just think the rating system is for data collection.
That’s why any evaluation system needs to stay transparent and actually deliver real improvements for users. This, in turn, strengthens the interaction between user and model, leading to ongoing improvements in the interface and experience.

Regenerate

The AI’s regenerate function creates fresh responses to the same prompt, really improving the user experience. Its perks include offering a range of options, giving users more control, letting you explore new ideas, and making it possible to track or compare previous answers.
For effective implementation, it’s important to keep the initial response, provide learnings on how the AI thinks, apply this to different situations, and give users parameters to maintain consistency. The regeneration feature makes interactions with AI more flexible, helping users get the best results.

Option to choose responses

The option to choose responses is a key tool for deepening and streamlining interactions between users and AI. It lets users pick from multiple AI answers. Benefits include promoting user-driven learning, boosting user control, and making the AI more adaptive.
By picking the best answer from a variety of options, users help the AI understand better and hit what they want more quickly. In turn, the AI gets a clearer sense of the user’s preferences and intent. Still, there are things to watch out for—like managing frustration when things don’t work, making sure there’s a way to track which answer gets picked, and keeping the user experience consistent. In short, this function gives users a strong tool to compare and choose AI responses, speeding up AI learning for more accurate, customized results—which helps build greater trust between AI and user.
Ran out of space—looks like there will be a Part 3.
Subscribe to 'haebom'
📚 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.
haebom@kakao.com
Subscribe