If you want a summary of Google I/O: – It's Gemini I/O, all they talk about is that. With demos and features that are repetitive and 90% of which will be available in several months for some (i.e., maybe never, we know Google). – The only interesting thing is Gemini 1.5
LLMS
-
Google Search integrating multi-step reasoning AI
By
–
More on search "Google will do the work for you" – Gen search will continue, and will roll out to everyone in US
– It will get multi-step reasoning!!! -
Full demonstration of Google’s AI agent
By
–
Full demo of Google’s AI Agent:
— AI Breakfast (@AiBreakfast) 14 mai 2024
Encodes video frames for real-time interpretation and conversational interactions pic.twitter.com/xgk6NxyAykFull demo of Google's AI Agent: Encodes video frames for real-time interpretation and conversational interactions
-
Google DeepMind Releases Gemini 1.5 Flash with 1M Token Context
By
–
Not enough info on Agents, feels like they closed it too fast. Now it goes to DeepMind and Gemini 1.5 Flash, a low latency 1M tokens available from today.
-
Potential for building AI agents inspired by Doordash
By
–
See Doordash. Looks like we will be able to build agents too!
-
AI Model Extends Context Window for Code Processing
By
–
Our apply model applies the change suggested by the chat and selected code block to the current file. We're working on extending the context window to handle 2000-3000 line files. Then, further improving accuracy and speed.
-
Faster AI Code Analysis Model Outperforms GPT-4
By
–
We've deployed a much faster apply model for files under 400 lines.
— Cursor (@cursor_ai) 14 mai 2024
It beats GPT4 and GPT4-o for short files on internal evals (but lags behind Opus).
(top is newer model, bottom is old) pic.twitter.com/wGL86PdgujWe've deployed a much faster apply model for files under 400 lines. It beats GPT4 and GPT4-o for short files on internal evals (but lags behind Opus). (top is newer model, bottom is old)
-
Gemini Pro context window expanded to 2M tokens
By
–
Gemini Pro got expanded from 1M to 2M context window
-
Double LLM Inference Speed with Medusa Speculative Decoding
By
–
Learn how to 2x LLM inference speeds with speculative decoding! We're introducing Medusa in our next release so you can accelerate inference by fine-tune open-source LLMs, whether or not you have labeled data. Join us on May 23rd at 10am PT! https://
pbase.ai/4ajjIxR -
Speculation on the future of Google’s conversational AI
By
–
So, a new conversational AI – separate app or Gemini? What are your bets?