step1_llm.py

"""
STEP 1 — A plain LLM call.

Goal: see the simplest possible interaction with an LLM,
and discover its main limitation: it can't act on the world.
"""
import os
import sys
sys.stdout.reconfigure(encoding='utf-8')
from dotenv import load_dotenv
from google import genai

load_dotenv()
# The client auto-reads the GEMINI_API_KEY environment variable.
client = genai.Client()

MODEL = "gemma-4-31b-it"

# A question the LLM CAN answer from training data
print("\n--- Q1: something it knows ---")
print(client.models.generate_content(
    model=MODEL,
    contents="In one sentence, what is retrieval-augmented generation (RAG)?",
).text)

# A question the LLM CANNOT answer — it has no tools, no web, no clock
print("\n--- Q2: something it CANNOT know ---")
print(client.models.generate_content(
    model=MODEL,
    contents="What is NVIDIA's latest stock price right now, and what is 22% of that?",
).text)

print("\n👉 Notice: the second answer is hallucinated or refused.")
print("   That's why we need TOOLS. On to step 2!")