New Secret GPT, the Pope and AI, Copilot Workspace — the top 3 AI news of the week

#1 — Ilya Sutskever left OpenAI

After half a year of silence and speculation, Ilya Sutskever, co-founder and chief scientist of OpenAI, left the company he had helped create. This is a huge loss for OpenAI because Ilya is the person who almost single-handedly developed the architecture of GPT and did much more for the industry. We are waiting to see Ilya at DeepMind or Anthropic.

#2 — Omni release

In other OpenAI news, the company presented its new GPT-4o (Omni) model (https://openai.com/index/hello-gpt-4o/). Apparently, this is the mysterious GPT that has been capturing the attention of Chatbot Arena users for the past month. “Omni” refers to the fact that it can handle text, speech, and video. The model is already available in ChatGPT (both paid and free!) and API. It has the same context as GPT-4 Turbo in 128k tokens but is completely multimodal. In blind tests, GPT-4o outperforms all competitors, although there are complaints that it is worse at prompt engineering than its predecessors. The new model will be rolled out in the voice assistant version of ChatGPT soon.

#3 — Google I/O conference highlights

On May 14, the largest conference for developers took place: Google I/O. Google traditionally uses the conference to tell and show how it will compete with the world of other products in the near future. Some interesting takeaways from this year:

— The new version of Gemini 1.5 Pro has 2 million tokens of context. That's a lot. So much that you can put the code of a not very small project into the context and work with it.

— Gemini will be integrated into all Google products and — most importantly — it will appear in Android 15 as an AI assistant called Project Astra that has the ability to see through the camera. Is Google going into competition with OpenAI's assistant?

— Gemini Nano, the smallest of the Gemini AI models, will be built into Chrome to run locally starting with version 126. Neural networks are going offline and on-device in leaps and bounds.

— Google’s AI chatbot for Android devices will feature an "Ask This Video" tool, which will allow "talking" to videos on YouTube — posing questions about the video that the chatbot will answer.

— Google also announced two new media models: Veo (sort of a lower-end SORA), which can generate high-def video, and Imagen 3, a text-to-image generator that may finally appear in Gemini Advance.

— For developers, Google showed an open source Firebase Genkit framework, which will allow you to embed AI in your apps on JavaScript/TypeScript and, soon, on Go.