AI Intelligence: Exploring the Future of Artificial Intelligence

Keeping Up with AI:
Gemma’s Here So
You Don’t Left Behind
Agadir
Mohammed Buallay

Mohammed Buallay
- Google Developers Experts for AI/ML
- Lead Data Architect and Reporting
- I love to travel and meet new people

Mohammed Buallay
- I started working on AI in 2017
- As senior project

Let’s meet with LLMs
What is Gemma?
Automate News Summary
Outlines

What is
an LLM?
[...] [...] [...]
[...]
0.02
0.03
0.9 0.01 0.0 …
Dogs Rain Drops Fish Wind …
and
cats
raining
It’s

,‫ﻧﺘﻌﻠﻢ‬ ‫ﻣﻨﻚ‬ ‫اﺣﻨﺎ‬ ‫و‬ ‫ﻣﻌﻠﻢ‬ ‫أﻧﺖ‬
Violets are blue,
Sugar is sweet,

‫ﻧﺘﻌﻠﻢ‬ ‫ﻣﻨﻚ‬ ‫اﺣﻨﺎ‬ ‫و‬ ‫ﻣﻌﻠﻢ‬ ‫أﻧﺖ‬
‫اﻟﻄﻴﺐ‬ ‫ﻓﻲ‬ ‫اﻟﻤﺜﻞ‬ ‫ﺑﻚ‬ ‫ﻧﻀﺮب‬
‫واﻟﻬﻤﺔ‬
‫اﻟﻘﻠﺐ‬ ‫ﻣﻦ‬ ‫وﻧﻘﻮﻟﻬﺎ‬ ‫ﺑﻚ‬ ‫ﻧﻔﺨﺮ‬ ‫ﻛﻠﻨﺎ‬
‫ﻓﻲ‬ ‫ﺑﺲ‬ ‫ﻣﻮ‬ ،‫اﻟﻔﻌﻞ‬ ‫ﻓﻲ‬ ‫ﻗﺪوﺗﻨﺎ‬
‫اﻟﻜﻠﻤﺔ‬
*

Modern LLMs can
be very large.
LLMs Explained

Classic Natural
Language Problems
LLMs Explained
Entity extraction Classification Summarization
Sentiment Analysis Translation Code Generation
…and more!

LLMs let us
prototype fast.
LLMs Explained

Explain this meme: Model: The meme plays of the
double meaning of “sprite”:
Model:
User:
Therefore, a game developer’s favorite drink is
Sprite because of their love for sprites in game
development, thus the drink’s name is a pun.
Sprite (the drink): popular lemon-lime soda
Sprite (2D bitmap image): often used to represent
characters, objects, etc. in game environments.
LLMs Explained

Model:
1. **Build a volcano.**
This is a classic science fair project that is both fun and educational. You
can learn about the science of volcanoes, earthquakes, and plate tectonics.
2. **Make a solar oven.**
This is a great project to learn about renewable energy and how to harness
the power of the sun. You can use your solar oven to cook food, dry fruit, or
even melt chocolate.
3. **Create a 3D printer.**
This is a more challenging project, but it is also very rewarding. You can
learn about 3D printing technology and design your own custom objects.
4. **Build a robot.**
This is a great project for those who are interested in engineering and
robotics. You can learn about basic electronics, mechanics, and
programming.
5. **Design a water filtration system.**
This is a project that can help you learn about the importance of clean
water and how to filter it. You can use your water filtration system to clean
water from a stream or river.
LLMs Explained
Give me five neat ideas for
science fair projects.
User:

Why are large language models
different?
LLMs are characterized by emergent
abilities, or the ability to perform tasks that
were not present in smaller models.
LLMs contextual understanding of human
language changes how we interact with
data and intelligent systems.
LLMs can find patterns and connections in
massive, disparate data corpora.
Search
Conversation
Content generation

What are some everyday tasks
that can be improved by AI?

Google is the industry pioneer in AI
Responsible AI
3,000
Researchers
7,000
Publications
Built & Tested
for Safety
Privacy in design
Upholds high scientific
standards
Accountable to People
Socially Beneficial
Avoid creating unfair
bias
2015
Google DeepMind
AlphaGo defeats Go
champion
2016
Google’s
DeepMind
helps detect
eye disease
2017
Google invents
Transformer
kickstarting LLM
revolution
2018
Google’s
groundbreaking
large language
model, BERT
2019
Text-to-Text
Transfer
Transformer
LLM 10B P
Model Open
Sourced
2020
Google LaMDA
Model Trained
to converse
2022
AlphaFold
predicts
structures of all
known proteins
2023
A conversational
AI Service
powered by
PaLM2
2024
Family of
multimodal LLMs &
products

GEMINI CAN
Understand
text, images,
audio and more

The Gemini Ecosystem
The most advanced AI from Google
For Developers
For Consumers
For Business and Enterprise
Models
Gemini API
(in Google AI Studio + ai.google.dev)
Gemini for Google Workspace
Gemini for Google Cloud
Gemini in Vertex AI
Gemini | app and web
Gemini in the Google App
Gemini in Gmails, Docs…
NotebookLM this deck
is about
But also Imagen,
Veo….
Unified SDK

Getting started with the
Gemini API
Train the Trainer in AI

AI Studio
aistudio.google.com
● Generate API Keys
● Create, test, and save prompts
● Customize models in minutes
● Generate starter code

ai.google.dev
REST API + new Client libraries for Python, Node, Java, and Go
Libraries
Libraries
Unified SDKs
…and more!

Vertex AI
Enterprise grade support.
Full MLOps (Examples: Model
evaluation, monitoring, registry)
Vertex AI
Check it out when you're
ready for enterprise-scale
production workloads.
Gemini API and Vertex AI
Both give access Gemini family
models.
Vertex AI
Cloud Production
ai.google.dev/docs/migrate_to_cloud
PlatfoEndpoints
rms
Platforms
Platforms
Now way easier with
the unified SDKs

Usage patterns

We’ll cover
● Basic prompting
● Code generation and execution
● Grounding with Google Search
● Function calling and structured outputs
● Vision (images, video) and audio understanding
● Image and video generation
● Tuning
Usage Patterns

Basic prompting

Write a poem about a magic backpack.
In a world of wonder, behold,
A backpack with tales yet untold.
Its fabric shimmers, alive with might,
A portal to realms, day and night.
aistudio.google.com

Write a poem about a magic backpack.
In a world of wonder, behold,
A backpack with tales yet untold.
Its fabric shimmers, alive with might,
A portal to realms, day and night.
...
aistudio.google.com

Generate a short tip about JavaScript. Good tips are short and
contain a code example showing the tip.
**Tip: Use Array.prototype.flat() to flatten an array of
arrays.**
```javascript
const array = [[1, 2], [3, 4], [5, 6]];
console.log(array.flat());
// Output: [1, 2, 3, 4, 5, 6]
```

Generate a short tip about JavaScript. Good tips are short and
contain a code example showing the tip.
topic: let and const
tip: Prefer using `let` and `const` to initialise variables,
instead of `var`, as `var` has no block scope or protection
against mutability.
e.g. const myVar = "EXAMPLE";
topic: arrow syntax
tip:

from google import genai
response = client.models.generate_content(
model="gemini-2.0-flash",
contents="Explain how AI works",
)
print(response.text)
# Okay, let's break down how AI (Artificial Intelligence) works.
It's a broad and complex field, so I’ll focus on the core
concepts and provide examples to make...

chat = client.chats.create(
model="gemini-2.0-flash"
)
response = chat.send_message(
"Hello, what should I have for dinner?")
# 'Here are some suggestions...'
"How do I cook the first one?")

Contents
Settings/Configs Text
Text
Text
Tools
Tools
Safety Settings
Safety Settings
Generation Config
Content of role: "user"
containing multimodal Parts:
Content of role: "model"
"user" content: a text Part
Tools specified by the caller
Safety setting configured by caller
Generation Config includes: temperature, Top P, Top K,
stop seq, max output tokens etc.
Generate Content Request

Candidate(s)
Feedback
Text
Block Reason
Safety Ratings
Candidate "Content"
note: you can ask for more than one content now
why the model stopped generating
feedback on the prompt
Finish Reason
Finish Message set if finish reason is present
Safety Ratings
Safety Ratings
how safe is the response
Generate Content Response
…

AI Studio
Endpoints
Prompting
https://ai.google.dev/docs/prompt_best_practices
● Chained prompts - make a plan, then execute it
● Context - Few shot prompts
● Generation parameters - Temperature, Safety settings, Top-P, TopK

Code generation

Code Generation
Code generation
Convert human language
requests to machine requests
Rapid prototyping for tools,
extensions, and applications
Generate code based on a
user prompt

Examples
Code generation
● Generate data
AI Studio example

Examples
Code generation
● Generate data
● Generate a SQL query
AI Studio example

Examples
Code generation
● Generate data
● Simulate Execution
AI Studio example

Examples
Code generation
● Generate data
● Quick coding help - Smart pair programer.
AI Studio example

Write a unit test for this Python function:
def is_diver_neutrally_buoyant(density_water, density_diver):
"""Check if a diver is neutrally buoyant."""
if density_water == density_diver:
return True
AI Studio example

import unittest
class IsDiverNeutrallyBuoyantTest(unittest.TestCase):
def test_diver_is_neutrally_buoyant(self):
self.assertTrue(is_diver_neutrally_buoyant(1000, 1000))
def test_diver_is_not_neutrally_buoyant(self):
self.assertFalse(is_diver_neutrally_buoyant(1000, 1010))
def test_invalid_input_types(self):
with self.assertRaises(TypeError):
is_diver_neutrally_buoyant("1000", 1000)
with self.assertRaises(TypeError):
is_diver_neutrally_buoyant(1000, "1000")

Examples
Code generation
● Generate data
● Quick coding help - Smart pair programer.
● Generating a full web application
● Integration in IDEs (ex: Copilot, Cursor, Continue, etc.)

Code Execution

Code execution
● Available via AI Studio and the Gemini APIs
● Gives Gemini the ability to write, execute, and fix Python code
● Sandboxed environment (no internet access)
● Can also output images!
Tuning goo.gle/gemini-api-cookbook-code-execution
AI Studio example

model = "gemini-2.0-flash",
contents = "Can you draw a graph of the first 50 values of the
fibonacci serie?",
config = types.GenerateContentConfig(
tools=[types.Tool(
code_execution=types.ToolCodeExecution
)]
)
)

model = "gemini-2.0-flash",
contents = "Can you draw a graph of the first 50 values of the
fibonacci serie?",
tools=[types.Tool(
code_execution=types.ToolCodeExecution
)]
)
)
# Okay, I can help you with that. I'll generate the first 50 Fibonacci
numbers and then create a graph plotting the index (from 1 to 50)
against the Fibonacci number. I will use python to generate the numbers
and then display a graph.

import matplotlib.pyplot as plt
def fibonacci(n):
fib_list = []
a, b = 0, 1
for _ in range(n):
fib_list.append(b)
a, b = b, a + b
return fib_list
fib_numbers = fibonacci(50)
plt.figure(figsize=(12, 6))
plt.plot(range(1, 51), fib_numbers, marker='o', linestyle='-')
plt.title('First 50 Fibonacci Numbers')
plt.xlabel('Index')
plt.ylabel('Fibonacci Number')
plt.grid(True)
plt.yscale('linear') # Or 'log' if you want to see the initial values better
plt.show()

Search Grounding

● Models have knowledge cut-offs
● Grounding with Google Search provides up-to-date
information
● Gives Gemini the ability to browse the web and cite sources
● 500 free RPD with 2.0 models
Grounding on Google Search
Search & IR goo.gle/gemini-api-cookbook-grounding
AI Studio example

● Instructions + Context + Question all in the prompt
● No need for RAG or fine-tuning
● Easy to implement
○ No extra code, just upload and ask.
Use the prompt's context window
Search & IR goo.gle/gemini-api-cookbook-pdf

contents="Who is Guillaume Vernade?",
config = {
'tools': [{
'google_search': {}
}]
}
)
# Guillaume Vernade, also known as Giom, is a tech enthusiast who
has worked in various roles, from neural networks and AI to Agile
coaching and video game production at companies like Ubisoft and
Google Stadia. Currently, he works at Google Deepmind as a
Developer Advocate, focusing on Generative AI with Gemini models.

pdf_file = client.files.upload(file="invoice.pdf")
contents=[
"Summarize this document",
pdf_file,
])
# This doc is an GCP invoice related to your Vertex AI usage…

Use the prompt's context window
Learning more
● gemini-2.0-flash: 1M tokens.
○ Remember: Everything is tokens.
○ Up to 2M+ tokens for Pro series
● Using context may be more flexible than Retrieval Augmented
Generation (RAG), or fine-tuning.
● More input context means higher latency
○ Use context caching to help reduce load times
https://blog.google/technology/ai/google-gemini-next-generation-model-february-2024
Search & IR

Search & IR
Deep-research
● Gemini agent built using grounding and function calling
● Plan first
● Then search, reason and report
Try deep research in the Gemini app

Search & IR
NotebookLM
● Understand complex info using your uploaded sources.
● Ask questions and get answers with citations.
● Generate summaries, study guides, and audio overviews.
● Works with Google Docs, PDFs, web pages.
NotebookLM

Function Calling

Function Calling goo.gle/gemini-fn-call-sql-github

Function Calling
https://github.com/GoogleCloudPlatform/generative-ai/tree/main/gemini/function-calling/sql-talk-app
SELECT ROUND((
COUNT(DISTINCT IF(returned_at IS NOT NULL, order_id, NULL)) /
COUNT(DISTINCT order_id)) * 100, 2)
AS return_rate
FROM thelook_ecommerce.orders
goo.gle/gemini-fn-call-sql-github

Function Calling
Function Calling
● Describe external functions to the model.
● The model may ask you to call the function to help it respond to
your queries (if not using automatic function calling).
● Endless possibilities for integrating external tools.
goo.gle/gemini-api-cookbook-function-callin
g

def dim_lights(brightness: float) -> bool:
"""Dim the lights."""
print(f”Lights are now set to {brightness:.0%}”)
return True
model='gemini-2.0-flash',
contents="Start the movie party"
tools=[dim_lights])

Function Calling
● Because function calling requires alternating turns, it's easiest
to use through chat.
● Enable "automatic function calling" when you start a chat, and
the Chat session will call the function(s) for you.
○ You don't have to use automatic function calling, it just
makes simple cases easier.
g

model = “gemini-2.0-flash”,
config = {
'tools': [set_light_values]
}
)
'Please turn down the lights about halfway.')

'Please turn down the lights about halfway.')
# Lights are now set to 50%.

Function Calling
● What happened? Use the chat history to find out.
● The chat history collects all the function calls and responses
that took place.
g

for content in chat.get_history():
part = content.parts[0]
print(content.role,":"))
if part.text:
print(part.text)
if part.function_call:
print("Function call: {", part.function_call, "}")
if part.function_response:
print("Function response: {", part.function_response, "}")

Contents
Tools
Text
Function Declaration
Function Declaration
Text
Function Calling interaction
Function Call
Function Response
Text
model may predict a function call
based on user content
model can understand the function response
and generate text
OR another function call
if one or more function declarations are
provided, function calling feature will turn on

Function calling - More Examples
Function Calling
● Wikipedia research aid
○ Integrates a search tool.
○ Uses the Gemini API inside the function call to summarize pages.
def wikipedia_search(queries:list[str]) -> list[str]:
...
https://ai.google.dev/docs/search_reranking_using_embeddings/

chat = model.start_chat(enable_automatic_function_calling=True)
model = 'gemini-2.0-flash',
config = {
"tools": [wikipedia_search],
"automatic_function_calling": {"disable": False}
# automatic_function_calling is enabled by default
}
)
res = chat.send_message("Explain how deep-sea life survives.")

chat = model.start_chat(enable_automatic_function_calling=True)
model = 'gemini-2.0-flash',
config = {"tools": [wikipedia_search]}
)
res = chat.send_message("Explain how deep-sea life survives.")
# Searching for "How do deep-sea creatures survive the extreme pressure?"
# Related search terms: ['Deep sea', 'Deep-sea community', 'Deep-sea fish']
# Fetching page: "Deep sea"
# Information Source: https://en.wikipedia.org/wiki/Deep_sea
# Fetching page: "Deep-sea community"
# Information Source: https://en.wikipedia.org/wiki/Deep-sea_community
# Fetching page: "Deep-sea fish"
# Information Source: https://en.wikipedia.org/wiki/Deep-sea_fish
# Searching for "How do deep-sea creatures survive the cold temperatures?"
# Related search terms: ['Deep-sea community', 'Deep sea', 'Deep-water coral']
# Fetching page: "Deep-water coral"

Function calling - Schema
Function Calling
● It's an OpenAPI schema, written as a protobuf.
● The protobuf-classes are available in the google.genai
client library.
● Functions can be written as JSON-compatible objects.
More examples:
https://ai.google.dev/gemini-api/docs/function-calling
https://ai.google.dev/tutorials/function_calling_python_quickstart

calculator = {
'function_declarations': [{
'name': 'multiply',
'description': 'Returns the product of two numbers.',
'parameters': {
'type': 'OBJECT',
'properties': {
'a': {'type': 'NUMBER'},
'b': {'type': 'NUMBER'}},
'required': ['a', 'b']}}]}

config = {
'tools': [set_light_values],
'automatic_function_calling': {'disable': True},
}

Structured output

Structured data
Structured output
● Structured data extraction.
● You can just ask the model to do it and return JSON.
goo.gle/gemini-api-cookbook-json-output
AI Studio example

contents=textwrap.dedent("""
Please return JSON describing the the people, places, things and relationships from this
story using the following schema:
{"people": list[PERSON], "places":list[PLACE], "things":list[THING], "relationships": list[RELATION
PERSON = {"name": str, "description": str, "start_place_name": str, "end_place_name": str}
PLACE = {"name": str, "description": str}
THING = {"name": str, "description": str, "start_place_name": str, "end_place_name": str}
RELATIONSHIP = {"person_1_name": str, "person_2_name": str, "relationship": str}
Here is the story:
""") + story)
AI Studio example

class Recipe(BaseModel):
recipe_name: str
recipe_description: str
recipe_ingredients: list[str]
class Cookbook(BaseModel):
cookbook_name: str
cookbook_description: str
cookbook_recipes: list[Recipe]
contents=”Create a cooking book about cookies, macarons and crème brulées.”,
config=types.GenerateContentConfig(
response_mime_type="application/json",
response_schema=Cookbook,
),
)
goo.gle/gemini-api-cookbook-json-output

Function calling - Structured data
Structured output
● Asking for JSON often works.
● Function calling lets you strictly describe the schema.
● With a strict description, we can strictly enforce that that's what
gets returned.

get_people = types.FunctionDeclaration(
name="get_people",
description="Get information about characters",
parameters=Person,
)
get_relationships = types.FunctionDeclaration(
name="get_relationships",
description="Get information about relationships between people",
parameters=Relationships
)
get_places = types.FunctionDeclaration(
name="get_places",
description="Get information about places",
parameters=Places
)

story_tools = types.Tool(
function_declarations=[get_people, get_relationships, get_places],
)

Function calling - Structured data
Function Calling
● Why?
○ Function calling doesn't return a string.
○ Function calling returns data-structures: You don't parse
text.
https://ai.google.dev/tutorials/structured_data_extraction

response = client.models.generate_content(f"""
Please add the people, places, things and relationships
from this story to the database:
{story}
""")
fc = response.candidates[0].content.parts[0].function_call
pprint.pprint(type(fc).to_dict(fc))

{'name': 'add_to_database',
'args': {'people': [{'description': 'A curious girl with wild hair of '
'midnight blue and eyes that shone like...,
'name': 'Luna',
...
'places': [{'description': 'Nestled between rolling hills and '
'sparkling rivers, where Luna and her '
'family lived.',
'name': 'Small town'}],
'relationships': [{'person_1_name': 'Luna', 'person_2_name': 'Elara',
'relationship': 'Mother-daughter'},
{'person_1_name': 'Luna', 'person_2_name': 'Orion',
'relationship': 'Father-daughter'}],
'things': [{'description': 'A dusty, forgotten backpack that Luna '
'found in the attic, which possessed a...',
'end_place_name': None,
'name': 'Backpack',
'start_place_name': 'Attic'},
...

Vision understanding

Spatial understanding
Multimodality
● Images and videos are just tokens in the input
● Can be used for instructions, context or query subject
● Also supports structured data outputs for visual content
goo.gle/gemini-2-spatial
AI Studio example

from google import genai
from google.genai import types
import PIL.Image
image = PIL.Image.open('/path/to/image.png')
client = genai.Client(api_key=GEMINI_API_KEY)
contents=["What is this image?", image])

Multimodality goo.gle/gemini-2-spatial

Image → structured data
Multimodality
● Generate text or structured data from images
goo.gle/gemini-api-cookbook-pdf

AI Studio
Endpoints
Images
Structured Data goo.gle/gemini-api-cookbook-pdf

Experimental spatial understanding
Multimodality
● Pointing at specific items
● Able to find them again in other pictures
● 3d understanding
goo.gle/gemini-2-spatial-3d

Multimodality goo.gle/gemini-2-spatial-3d

Video understanding
Multimodality
● Search within videos
● Extract information and text
● Structure information
● Analyse key moments
● Soon: Youtube integration
goo.gle/gemini-api-cookbook-video
AI Studio example

Audio understanding
goo.gle/gemini-api-cookbook-audio
AI Studio example

Live API

Live API
Live API
● Low-latency bidirectional voice and video interactions
● Outputs natural, human-like voice (English only)
● Multi-tools integration (but no automatic function calling)
● Experimental, limited to 10mn
● Uses websockets instead of REST
goo.gle/gemini-api-cookbook-liveapi
AI Studio example

Multimodal generation

Imagen
Multimodal output
● Our highest quality text-to-image model
● Generates stunning, detailed, artifact-free images.
● Understands complex prompts and diverse styles.
● Superior text rendering for creative uses.
● Built with robust safety and watermarking (SynthID).
● Some extra options when using Vertex AI (negative prompt,
image2image)
● Paid tier only (0.03$/image)
goo.gle/gemini-api-cookbook-imagen
Prompting guide

Image-out
Multimodal output
● Currently in early-access
● Gemini can generate images as well as text
● Optimized for multi-turn conversations

Audio-out
Multimodal output
● Gemini can generate audio directly (like Live API)
● Optimized for multi-turn conversations

Veo
Multimodal output
● State-of-the-art realistic video generation.
● Produces high-quality 4K video output.
● Advanced motion and physics simulation.
● Precise camera control and shot styling.
● Superior realism and artifact reduction.

Tuning

Tuning
● If you aren’t able to get good results with:
○ Prompting
○ Retrieval (RAG)
○ Tools and function calling
○ Search grounding
…then try fine-tuning.
Tuning
https://ai.google.dev/tutorials/tuning_quickstart_python

Colab: Inference images with Gemma3

AI Intelligence: Exploring the Future of Artificial Intelligence

More Related Content

Similar to AI Intelligence: Exploring the Future of Artificial Intelligence (20)

Recently uploaded (20)

AI Intelligence: Exploring the Future of Artificial Intelligence