#259 - Dots, Sonnet, Self-Safety, Rogue AI
Summary
Hosts Andrey Krennikov and Jeremy Harris discuss OpenAI’s new GPT-6 Sol and Anthropic’s Sonnet 5.5, noting their cost-efficiency amidst paused frontier releases. They analyze OpenAI’s launch of "Dots," a consumer agentic assistant, and Meta’s camera-free AI glasses. The episode also covers AMD’s $8.2 billion acquisition of World Labs and Anthropic’s IPO filing, which highlights catastrophic AI risks and a governance structure granting founders significant voting control.
Topics discussed
Intro, host banter, and Langfuse sponsorship
Listener feedback and podcast format adjustments
Model release pace and OpenAI Astra 6.1 cancellation
Anthropic Sonnet 5.5 release and cost comparisons
OpenAI Dots and Google Gemini Spark agent launches
Rogue AI incidents and the rise of agent credentialing
OpenAI ChatGPT updates and B2C user experience focus
Meta Muse integration and smart glasses privacy debate
Google Gemini 4 Pro leaks and frontier race status
AMD acquires Fei Fei Li's World Labs for $8.2B
Business updates and financial performance details
Anthropic governance changes and founder voting power
New lightweight decision-making model architectures
Sponsorships: Optimize workflow tool and meal delivery
Jevons paradox and open-source JEV alternatives
White House AI safety guidelines and antitrust concerns
Whistleblowers, employee morale, and regulation barriers
OpenAI security incidents and agent coordination failures
Cybersecurity benchmarks and model hacking capabilities
GLM 5.3 cyber capabilities and China open-source impact
Recursive Self-Improvement (RSI) and optimization strategies
Listen ad-free on Castria