# “How AI APIs Work: Response Types, Hallucination, and How to Handle Them”

# What Are AI APIs?

AI APIs are **cloud-based interfaces** that allow us to **access and use artificial intelligence models** — like language, vision, or speech models — without needing to train or host them on their own servers.

Think of AI APIs as a teacher who knows it all and can reply with text, images, voice, and much more — all instantly. And the best part? We don’t need to know how they do it; we just ask and get the answer.

## Where they’re used

AI APIs are everywhere! From chatbots that talk like humans, to tools that write your blogs or ads for you. They power virtual assistants, create stunning images and lifelike voices, tutor students with personalized lessons, and even help developers write and fix code — all in the blink of an eye.

Nowadays, almost every web or mobile app is evolving by integrating AI APIs to stay ahead with today’s technology. Developers use AI APIs to add smart features like chatbots that understand natural language, content generators that write automatically, image and voice creators, personalized learning assistants, and code helpers. These APIs let apps offer powerful AI capabilities without building complex models from scratch.

## Types of Responses from AI APIs

### 1\. **Text Completion**

This is the classic mode, where you send a prompt and get a plain text reply.

```python
from google import genai

# Initialize Gemini model
model = genai.GenerativeModel("gemini-1.5-pro")

# Send prompt
response = model.generate_content("Write a short poem about the moon.")

# Print plain text reply
print("Plain text reply:")
print(response.text.strip())
```

### 2\. **Function Calling / Tool Use**

Some AI APIs support structured responses where the AI can return JSON objects to trigger actions or call functions in your app.

```python
from google import genai
import json

# Initialize Gemini model
model = genai.GenerativeModel("gemini-1.5-pro")

# Send prompt requesting structured response
response = model.generate_content(
    "Return JSON to call a function named 'add_numbers' with arguments a=5 and b=7. No explanation."
)

# Parse text as JSON
result_json = json.loads(response.text.strip())

# Print the structured result
print("Simulated function call JSON:")
print(json.dumps(result_json, indent=2))

# Example: You could then trigger your add_numbers function with this data.
```

### 3\. **Structured Output**

You ask the model to return structured data (usually JSON) based on pattern or schema-but its just data no action

```python
def generate_content(prompt: str):
    load_dotenv()
    api_key = os.getenv("GEMINI_API_KEY")
    client = genai.Client(api_key=api_key)
    response = client.models.generate_content(
    model="gemini-2.0-flash",
    contents=prompt,
    config={
        "response_mime_type": "application/json",
        "response_schema": CourseResponse, #Here CourseResponse pydantic model
    },)
    try:
        parsed = json.loads(response.text)
        return CourseResponse(**parsed)
```

## Why Do Hallucinations Happen in AI APIs?

**Hallucination** refers to when an AI confidently produces **false, misleading, or completely made-up responses**. This can be a major issue in real-world applications.

### Examples of Hallucination

1. **Fake Citations**
    

> “According to Smith 2022 in *The AI Guidebook*…” (book doesn’t exist)

2. **Bad Code**
    

> Returns Python code that imports a nonexistent module

3. **Wrong Math**
    

> Calculates `23 × 5 = 135` (because it "sounded right")

### Why Do AI Models Hallucinate?

AI models sometimes produce false or made-up information, which is known as hallucination. This happens for several reasons:

1. **They predict text, not facts.**  
    Language models are trained to predict the next word based on patterns, not verify truth.
    
2. **Vague or unclear prompts.**  
    If your input is too short or missing detail, the model fills in the blanks with guesses.
    
3. **Outdated or limited knowledge.**  
    Many models don’t have real-time access to the internet, so their knowledge may be outdated.
    
4. **No access to tools.**  
    Without plugins or API access, the model can't look up facts, calculate, or fetch real data.
    
5. **High temperature settings.**  
    A higher temperature makes responses more creative, but also increases the chance of hallucination.
    

### What is Temperature in AI?

**Temperature** is a setting that controls how creative or random the AI’s responses are.

* **Low temperature (e.g., 0–0.3):**  
    The AI gives more focused, predictable, and accurate answers. Good for facts and code.
    
* **High temperature (e.g., 0.7–1.0):**  
    The AI becomes more creative, diverse, and expressive. Good for storytelling or brainstorming — but more likely to make things up (hallucinate).
    

### How to Handle and Reduce Hallucination

1. **Write clear prompts**  
    Be specific. The more detail you give, the less the model needs to guess.
    
2. **Use system messages**  
    Set the model’s role, like: “You are a helpful assistant. Only respond with facts.”
    
3. **Validate responses**  
    Check outputs before using them — especially for structured data. Use tools like `json.loads()` or Pydantic.
    
4. **Post-process results**  
    Add logic to catch and filter out fake data, bad links, or unsupported answers.
    
5. **Use function calling or structured output**  
    Instead of asking for text, have the model return specific arguments that your system understands.
    
6. **Lower the temperature**  
    Set `temperature=0` for more accurate, fact-based responses.
    
7. **Use real-time tools or APIs**  
    Connect the AI to your own tools, databases, or search functions to avoid guesses.
    
8. **Have a human in the loop**  
    For critical use cases, let a human review the AI’s response before it’s published or used.