How to use OpenAI Decisions API for image judgments and real time UI

TL;DR

The video introduces the OpenAI Decisions API which can judge both text and images, and its use with a real time intelligent UI in ChatGPT. It explains how to set inputs, interpret outputs such as probabilities and categories, and demonstrates several demo scenarios including image damage checks, receipts classification, and risk scoring. It also covers pricing and practical business applications.

Transcript

Hi, hello. I'm Keto. This channel introduces the latest information and use cases for generative AI. Please make sure to subscribe to the channel. And this time, OpenAI has released a new feature called the Decisions API. So, what is this Decisions API? It is almost identical to the tool I introduced the other day called Jev, an AI you can... Read More

Key Insights

• The Decisions API can process both text and images, returning structured judgments rather than narrative text.
• Outputs can be probabilities, binary classifications, or scoring based on fixed criteria, enabling fast triage of large data sets.
• Pricing information shows free output and 0.10 per million tokens for input, making it cost effective for large scale use.
• Demo examples demonstrate damage detection in products, categorization of receipts, and grading of hotel room cleanliness to show practical utility.
• The tool can flag scam messages and identify urgent inquiries, supporting customer support workflows.
• Inputs can be text, images, or mixed data, with questions asked about the inputs to guide the model’s judgments.
• Three basic patterns of judgments are probability, selection, and scoring, which cover common business needs.
• The Intelligent UI in ChatGPT adds real time, visual interaction to judgments, enhancing user experience and decision speed.

Questions & Answers

Q: What is the OpenAI Decisions API and what can it judge?

The OpenAI Decisions API is described as an AI for making judgments rather than generating text. It can process both text and images, providing outputs such as probabilities, category selections, or scores based on a fixed rule set. This enables rapid triage and classification across large item sets.

Q: How does the Decisions API handle image judgments in the demo?

In the demo, the API analyzes product images for damage, categorizes items like mugs and boxes, and outputs a confidence level or a binary assessment such as damaged or not damaged. It can also identify non damage scenarios and still provide useful categorization such as product type.

Q: What are the pricing basics for the Decisions API?

The video explains that the output from the Decisions API is free, while the input cost is 0.10 per million tokens for text related processing. This suggests a cost effective model for large scale inference, especially when handling many items across different formats.

Q: What kind of outputs can you expect from the API for text questions?

For text inputs, the API can categorize inquiries, assign them to appropriate departments, or identify actions like urgency or refund needs. It produces structured outputs such as category labels or urgency scores, enabling automation and routing without generating long text responses.

Q: What is Intelligent UI in ChatGPT and how does it relate to Decisions API?

Intelligent UI is described as a real time visual interaction layer on ChatGPT. It complements the Decisions API by providing immediate, visual feedback and interaction with judgments, helping users adjust inputs and observe results in a live, interactive way.

Q: What are some sample business use cases mentioned for the Decisions API?

Sample use cases include checking product damage and categorizing expenses from receipts, grading hotel room cleanliness, and ranking safety on construction sites. The examples illustrate how decisions can be made quickly with image analysis and text sorting to support operations and customer support.

Q: How does the Decisions API handle mixed inputs and questions?

The API accepts both text and image inputs, and users can ask questions about them. This allows multi modal judgments such as assessing conditions in a photo while also extracting relevant text, enabling complex workflows that combine visual and textual information.

Q: What potential cost or efficiency benefits are highlighted in the video?

The video emphasizes high speed and low per item costs, with fast judgments on thousands of items at once. It shows how fixed criteria produce precise outputs, enabling automation of routine decisions, reducing manual effort, and potentially lowering operational costs in areas like customer support and logistics.

Summary & Key Takeaways

• The video explains the Decisions API as a fast, rule based judgment tool that can handle text and images, outputting probabilities or categorical results rather than long text. It emphasizes practical demos to show how input and output parameters work in real scenarios.

• It demonstrates a demo site with seven image judgment cases, including product damage, expense categorization, and safety checks. It also shows how text judgments can sort inquiries, assign refunds, and flag immediate responses, highlighting speed and cost efficiency.

• A section highlights the intelligent UI in ChatGPT that enables real time visual interaction, with use cases and display settings. The narrative ties Decisions API usage to concrete business workflows like help desks, QA, and operations, making the features feel ready for adoption.