Gemini, Google's Large Language Model
summary
Gemini is a multimodal LLM with sizes Nano, Pro, Ultra and a Flash flavor that streams responses fast and supports a 1 million token context (examples: one hour video, 11 hours audio, 30k lines of code). Demos show image/video parsing (found an amethyst geode), transcript search returning the exact Apollo 11 timecode, streaming token output, RAG via embeddings plus a vector DB, function-calling to invoke a weather API, and Java integration using LangChain4j and the Gemini Java SDK. Also available as local Gemma 2B and 7B open models running in Docker.
AI-written summary · about
chapters (15)
- 0:00Google's Gemini intro
- 1:26Transformers revolution
- 3:13Gemini sizes and features
- 7:40Gemini web app demo
- 13:34Gemini multimodal demo
- 20:04Gemma open models
- 22:35Using Java with large language models
- 26:52Java chat and streaming
- 32:13Java image interactions
- 33:17Prompt templates and sentiment
- 35:05Data extraction
- 36:54Retrieval augmented generation in Java
- 42:38Function calling
- 47:42Run Gemma locally
- 50:11Conclusion
related talks
54mGemini, Google's Large Language Model
51mGemini, Google's Large Language Model
19mGemini 2.0: Beyond the Basics
16mOpen models in the Gemini era
56mBuilding Multimodal Apps with Gemini 2 0 and Gemini SDK
18mIntroduction to Gemini APIs and AI Studio
01h 01mBuilding Multimodal Apps with Gemini 2.0 and Gemini SDK
15mBuilding with the Gemini API and AI Studio