Gemini 2.0 per sviluppatori (Milano 20mar2025)

Deck for many succulent Gemini 2.0 demos in /dev/cloud day Milan 2025.

Transcript

1. per sviluppatori Milano, 20 Marzo 2025 Riccardo Carlesso Developer Advocate
2. Imagine an AI that can read, hear, see, and respond in realtime
4. Gemini is an umbrella brand used in Google products that utilize the Gemini model
5. Hello to
13. Features across models
14. Features across models Feature Gemini 2.0 Flash Gemini 2.0 Flash-Lite Gemini 2.0 Pro Experimental Release status Generally available Generally available Available today Multimodal inputs ✅ ✅ ✅ Text output ✅ ✅ ✅ Image output Coming soon… ❌ Coming soon
15. Features across models Feature Gemini 2.0 Flash Gemini 2.0 Flash-Lite Gemini 2.0 Pro Experimental Release status Generally available Generally available Available today Multimodal inputs ✅ ✅ ✅ Text output ✅ ✅ ✅ Image output ✅ !!! ❌ Coming soon
16. Features across models Feature Gemini 2.0 Flash Gemini 2.0 Flash-Lite Gemini 2.0 Pro Experimental Release status Generally available Generally available Available today Multimodal inputs ✅ ✅ ✅ Text output ✅ ✅ ✅ Image output ✅ !!! ❌ Coming soon
17. gemini-2.0-flash A GA model on: • Gemini Developer API, on Google AI Studio • Gemini API, on Vertex AI Studio
18. gemini-2.0-flash-thinking-exp-01-21 An experimental model trained to generate the thinking process for stronger reasoning across math and science
19. Performance Multimodal Live API Unified SDK Native Tool Use Native Image & Audio Output (preview) Spatial Understanding
20. Enterprise Vertex AI Studio Google Cloud APIs cloud.google.com/vertex-ai Consumers Gemini gemini.google.com Developers Google AI Studio Google APIs aistudio.google.com ai.google.dev/gemini-api
21. Rio de Janeiro Demo1 Rio de Janeiro - chapterize
22. Performance
23. Gemini Flash 2.0 offers 2x the speed of Gemini 1.5 Pro Stronger performance on multimodal, text, code, video, spatial understanding and reasoning
27. Rio de Janeiro Demo2 Demo2: In AI Studio, show Gemini 2.0 Flash Thinking Mode with pool image and this prompt: How do I use three of the pool balls to sum up to 30?
29. 6 + 11 + 13 = 30
30. Unified SDK
32. Unified interface to Gemini 2.0 (and 1.5) • Gemini Developer API on Google AI Studio • Gemini API on Vertex AI The new Google Gen AI SDK
33. Gemini Developer API on Google AI Studio client = genai.Client( api_key=your-gemini-api-key) response = client.models.generate_content( model="gemini-2.0-flash-001", contents="Why is the sky blue?")
34. Gemini Developer API on Vertex AI client = genai.Client( vertexai=True, project=your-google-cloud-project, location="us-central1") response = client.models.generate_content( model="gemini-2.0-flash-001", contents="Why is the sky blue?")
35. Using environment variables Gemini Developer API on Google AI Studio export GOOGLE_API_KEY='your-api-key' Gemini Developer API on Vertex AI export GOOGLE_GENAI_USE_VERTEXAI=true export GOOGLE_CLOUD_PROJECT='your-project-id' export GOOGLE_CLOUD_LOCATION='us-central1' client = genai.Client() Common client initialization
36. Native Tool Use
37. Google Search Tool Ground model responses in Google Search results For more accurate, up-to-date, and relevant responses
38. Why grounding? Ask an LLM: ⚛Explain Einstein’s Relativity 🪐How many moons does Saturn have? ⌚What time is it today? 🌤What is the weather like in Milan?
39. Google Search Tool google_search_tool = Tool(google_search=GoogleSearch()) response = client.models.generate_content( model="gemini-2.0-flash-001", contents="How’s the weather like today in Milan?", config=GenerateContentConfig(tools=[google_search_tool]) )
40. Rio de Janeiro Demo3 Demo3: In Vertex AI Studio, show this prompt without and with Google Search grounding and also show the code: Che tempo fa oggi a Milano? [Specifica in formato JSON]
41. Code Execution Tool Model generates and runs Python code Useful for applications that benefit from code-based reasoning (e.g. solving equations)
42. Code Execution Tool code_execution_tool = Tool(code_execution=ToolCodeExecution()) response = client.models.generate_content( model="gemini-2.0-flash-001", contents="What is the sum of the first 50 prime numbers?", config=GenerateContentConfig( tools=[code_execution_tool], temperature=0))
43. Rio de Janeiro Demo4 Demo4: In Google AI Studio, show this prompt without and with Code Execution: What is the sum of the first 50 prime numbers? (should be 5117)
44. Automatic Function Calling Submit a Python function for automatic function calling (instead of submitting an OpenAPI specification of the function)
45. Automatic Function Calling def get_current_weather(location: str) -> str: """Example method. Returns the current weather. Args: location: The city and state, e.g. San Francisco, CA """ weather_map: dict[str, str] = { "Barcelona": "sunny", "Paris": "foggy", "Milan": "raining", "Rome": "hot", "London, UK": "rainy and dark", } return weather_map.get(location, "unknown")
46. Manual Function Calling (before)
47. Automatic Function Calling response = client.models.generate_content( model="gemini-2.0-flash-001", contents="What is the weather like in Milan?", config=GenerateContentConfig( tools=[get_current_weather], temperature=0))
48. Maps Function Calling within Gemini Che pizzerie ci sono a Milano? Rimaniamo in zona Isola (entro 1km da via Confalonieri). Siamo a piedi. -- Ora trova il numero di telefono di quei ristoranti Ora aggiungi la distanza dall’ufficio di Google (in via Confalonieri) -- Ok allora per questi ristoranti per favore dammi numero di telefono, indirizzo, breve descrizione in forma tabulare. Su gemini.google.com 48 Dati pizzerie Isola per gente rumorosa
49. Spatial Understanding
50. Improved accuracy on 2D and 3D spatial understanding Spatial Understanding https://aistudio.google.com/starter-apps/spatial
51. Improved accuracy on 2D and 3D spatial understanding Spatial Understanding https://aistudio.google.com/starter-apps/spatial
52. Rio de Janeiro Dem Demo5: Open https://aistudio.google.com/starter-apps/spatial and go through some scenarios: 1. Animal shade // animal name in Italian with emoji 2. Fish // specific food name of in italian
54. Multimodal Live API
55. Enables low-latency, two-way interactions → Input: text, audio, and video ← Output: audio and text Multimodal Live API
56. • Multimodality – model can see, hear, speak • Low-latency – for realtime interaction • Memory – model remembers the session • Tools – Function calling, code execution, and Google search Multimodal Live API–key capabilities
57. Rio de Janeiro Dem Demo6: First, show that Live API is available in Google AI Studio. Then, show 1-2 of these starter apps: cd ~/git/pvt-gemini20
58. Ricc Demo
59. Native Image & Audio Output (preview)
60. Gemini 2.0 introduces native image generation and text-to-speech capabilities Enables image generation / editing and expressive storytelling Native Image & Audio Output (preview)
63. Rio de Janeiro Dem Demo7: Image output -> AI Studio -> Seleziona Gemini Video > Visual Story
64. Gemini2.0: text+image (story) Shrek in Milan: A Digital Story
65. https://www.youtube.com/shorts/sA Ah4aI7ZA8 Veo: text to video
66. https://www.youtube.com/shorts/sA Ah4aI7ZA8 Veo: text to video “Shrek and Fiona are doing an Ironman in Japan: Shrek is cycling in a trisuite while Fiona is getting out of the water right now. The camera fades out, then the view goes to above. On a distance, donkey is taking pictures of the green couple, on top of his big red dragon.“
67. Veo: text+image to video “Person becomes Shrek in front of Milan Duomo square, then eats a panettone”
68. Veo: text+image to video “This code enters a rocket ship and flies to the colorful clouds of Google Cloud!
69. Veo: text+image to video “This code enters a rocket ship and flies to the colorful clouds of Google Cloud!
70. Veo app “mosaic” con Streamlit $ cd ~/git/genai-googlecloud-scripts/22-gemini20/ && make app => http://localhost:8501/
71. goo.gle/gemini2 goo.gle/multimodal-live-api Keep exploring
72. Thank you! Riccardo Carlesso Developer Advocate at Google @palladius Ricc.rocks http://linkedin.com/in/riccardo-carlesso https://speakerdeck.com/palladius