LLM Bootcamp Notes — Part 3: UX for Language User Interfaces
The LLM Bootcamp series by FullStackDeepLearning offers great insights into the world of Generative AI by taking a very structured approach to the topic. The series goes from introducing LLMs all the way upto approaches and recommendation for production grade LLMs & Generative AI solutions.
I have consolidated my learning notes in a series of 4 articles, distilling to the 4 core focus areas for LLMs from the Bootcamp series — This part 3 article is on UX for Language User Interfaces.
UI PRINCIPLES
The main focus of any UI principle is to make a “human centred design” — This ivloves designing an user interface, both analog and digital, by making it easy, efficient and intuitive for a human mind to understand and execute. There are a few baseline pointers here:
- Affordances: Possible actions offered by an object. Eg: Push & Pull of a door
- Signifier: Cues that indicate affordances . Eg: Label to say whether to push or pull the door
- Mapping & Feedback — Relationship between control and their effect. A good design should have intuitive mapping to reduce cognitive load
UI Considerations:
- Design for scanning, not reading — the websites/apps should be able to give the most information in minimum time
- Make actionable things unambiguous, instinctive and conventional
- Options are needed, but it should be not too many and the presnted ones must be clearly distinguishable from one another
AI product considerations:
Is AI a good solution to for this problem? This is an eternal question that makes or breaks a project. A thumb-rule framework that help with making the decision is this:
- No AI — If the performance of AI on the task is bad and if they danger levels are high. Eg: Full self-driving cars were long touted as the next big thing, but it is still not achieved because of the AI performance and the high-danger levels associated.
- Assisted AI — If the AI performance is satisfactory/near-human level and the danger levels are manageable. Eg: Predicting the house price in the neighbourhood, where the AI gives a recommendation of the price by taking in the factors associated. The power of AI here is giving assistance/recommendations but does not perform actual buying/selling transactions.
- Automated AI — If the AI performance beats humans and if the danger levels are low to medium. For Eg: Movie & Song recommendation, route prediction etc
Language UI (LUI) PATTERNS
Language user interfaces play an important role in creating a successful AI product, especially language based AI products. As always, the interface should be simple and intuitive to use. Additionally, a LUI must also be able to collect natural feedback from the user for the quality and relevancy of the response given. There are a many popular Generative AI Language solutions — with them some LUI patterns that currently exists:
- Click to complete (OpenAI playground) — the user has to explicitly click the generated answer in order to accept & populate it. This ensures whether the user is satisfied the answer or not
- Auto-complete (Copilot) — In Copilot the code generation happens in an auto-complete mode where the code is written
- Command palette (replit) — the question must be posed explicitly to the AI, otherwise it won't interfere or produce any suggestions on your work
- Chat interface (ChatGPT) — A QnA based system where the AI responds with 1 answer to 1 question. This happens in a sequential manner.
Here are a few guiding questions for building a good LUI:
- What is the boundary — defining the limit/scope of the product
- How high is the accuracy requirement
- Is latency a key factor
- Are users encouraged to give feedback
Footnotes
To read the part 1 article of this series — Click here
To read the part 2 article of this series — Click here
To read the part 4 article of this series — Click here
Note — The article is a distilled consolidation of my understanding of the topic. If you find any conceptual errors, please leave a feedback so that I can fix it. Cheers!
References:
