#244 - GPT-5.5 Instant, Grok 4.3, OpenAI vs Musk
Summary
Hosts Andrei Kurenkov and Jeremy Harris discuss the real-world impact of Mythos on cybersecurity, defending its significance against skeptics. They review OpenAI’s GPT-5.5 Instant, noting its cyber risks and a quirky "goblin" training anomaly. The episode also covers XAI’s cost-effective Grok 4.3, Mistral’s unified Medium 3.5 model, and Anthropic’s new agent features like "Dreaming" for self-improvement.
Topics discussed
Intro, Box AI Agent sponsorship, and MIFOs discussion
Mythos paper analysis and bioweapon risks
OpenAI o1 model capabilities and training quirks
Grok 3, Mistral, and Cohere model updates
AI agents, dreaming, and Suno music licensing
OpenAI lawsuit, Elon Musk, and Sam Altman drama
XAI compute deals, Palantir, and FIS partnership
Data center sales, bank debt, and financial risks
Anthropic funding, China AI, and interpretability
NVIDIA GB200, Pentagon contracts, and safety
NSA cyber offense and safety adapter training
Recursive agents, Hamiltonian dynamics, and chess
Outro and listener questions
Listen ad-free on Castria