Google DeepMind recently released a report about its new Gemini family of models: "Gemini: A Family of Highly Capable Multimodal Models". This video summarises its content.
Timestamps:
00:00 - How does Gemini compare to GPT-4?
01:07 - Model Architecture
02:13 - Training Infrastructure
06:14 - Training Dataset
06:52 - Evaluation and comparison to GPT-4
09:01 - Use of long context (32K tokens)
09:50 - AlphaCode 2
10:07 - Multimodal evaluations
11:57 - Gemini helps a user cook a Veggie Omelette
12:54 - Gemini approach to responsible deployment
The tech report can be found here: https://storage.googleapis.com/deepmi...
An overview of the use of self-supervised learning in Computer vision.
Topics: #gemini #foundation #frontier
References for papers mentioned in the video can be found at
https://samuelalbanie.com/digests/202...
For related content:
Twitter: / samuelalbanie
personal webpage: https://samuelalbanie.com/
YouTube: / @samuelalbanie1