#210 - Claude 4, Google I/O 2025, OpenAI+io, Gemini Diffusion | Last Week in AI

In this episode of the Last Week in AI podcast, Andrey Kurenkov and Jeremie Harris discuss the latest AI news, focusing on Google's I/O 2025 announcements, including AI mode in Google Search, Project Mariner, VO3, Imagine 4, and real-time speech translation in Google Meet, as well as GitHub's new AI coding agent and Mistral's DevStral. They also cover OpenAI's acquisition of IO, its planned data center in Abu Dhabi, and LM Arena's funding round. The episode further delves into research advancements like Gemini Diffusion, Chain of Model Learning, and Seek in the Dark, and concludes with policy and safety updates, including OpenAI's response to California's Attorney General and Anthropic's activation of AI Safety Level 3 protections.

Outlines

Part 1: Claude 4 and Google's AI Search

Part 2: Google's AI Tools and Coding Agents

Part 3: Security and Funding

Part 4: Model Performance and Training

Part 5: Improving Model Performance

Part 6: AI Safety and Conclusion

Sign in to continue reading, translating and more.

Open full episode in Podwise

#210 - Claude 4, Google I/O 2025, OpenAI+io, Gemini Diffusion

Last Week in AI

Part 1: Claude 4 and Google's AI Search

Introduction to the AI News of the Week and Claude 4 Announcement

Claude 4 Performance and Agentic Capabilities

Google's AI Mode in Search and Project Mariner

Part 2: Google's AI Tools and Coding Agents

Project Mariner Updates and Google's V03

Google's Imagine 4 and AI Coding Agents

Coding Agents and OpenAI's Acquisition of IO

Part 3: Security and Funding

OpenAI's Data Center in Abu Dhabi and Security Concerns

LM Arena's Funding and the State of AI Servers

Part 4: Model Performance and Training

Google's AI Advancements and Meta's Llama Delay

Meta's Struggles and Google's Gemini Diffusion

Chain of Model Learning for Language Model

Part 5: Improving Model Performance

Seek in the Dark and Two Experts Are All You Need

Defending Gemini Against Indirect Prompt Injections

How Fast Can Algorithms Advance Capabilities?

Part 6: AI Safety and Conclusion

OpenAI's Response to California's Attorney General and Anthropic's AI Safety Level 3 Protections

Conclusion

#210 - Claude 4, Google I/O 2025, OpenAI+io, Gemini Diffusion

Last Week in AI

Part 1: Claude 4 and Google's AI Search

00:10Introduction to the AI News of the Week and Claude 4 Announcement

Introduction to the AI News of the Week and Claude 4 Announcement

04:49Claude 4 Performance and Agentic Capabilities

Claude 4 Performance and Agentic Capabilities

10:01Google's AI Mode in Search and Project Mariner

Google's AI Mode in Search and Project Mariner

Part 2: Google's AI Tools and Coding Agents

15:21Project Mariner Updates and Google's V03

Project Mariner Updates and Google's V03

21:26Google's Imagine 4 and AI Coding Agents

Google's Imagine 4 and AI Coding Agents

27:35Coding Agents and OpenAI's Acquisition of IO

Coding Agents and OpenAI's Acquisition of IO

Part 3: Security and Funding

36:06OpenAI's Data Center in Abu Dhabi and Security Concerns

OpenAI's Data Center in Abu Dhabi and Security Concerns

41:15LM Arena's Funding and the State of AI Servers

LM Arena's Funding and the State of AI Servers

Part 4: Model Performance and Training

48:02Google's AI Advancements and Meta's Llama Delay

Google's AI Advancements and Meta's Llama Delay

54:55Meta's Struggles and Google's Gemini Diffusion

Meta's Struggles and Google's Gemini Diffusion

1:00:09Chain of Model Learning for Language Model

Chain of Model Learning for Language Model

Part 5: Improving Model Performance

1:08:24Seek in the Dark and Two Experts Are All You Need

Seek in the Dark and Two Experts Are All You Need

1:15:38Defending Gemini Against Indirect Prompt Injections

Defending Gemini Against Indirect Prompt Injections

1:23:34How Fast Can Algorithms Advance Capabilities?

How Fast Can Algorithms Advance Capabilities?

Part 6: AI Safety and Conclusion

1:31:11OpenAI's Response to California's Attorney General and Anthropic's AI Safety Level 3 Protections

OpenAI's Response to California's Attorney General and Anthropic's AI Safety Level 3 Protections

1:42:54Conclusion

Conclusion