#241 - Opus 4.7, Muse Spark, GPT-5.4-Cyber, HY-World 2.0
Apr 23, 2026
More Technology podcasts
Summary
Hosts Andrey Krenkov and Jeremy Harris review a product-heavy week in AI, focusing on Anthropic’s Claude Opus 4.7 and its detailed safety system card. They discuss Meta’s MuseSpark model and OpenAI’s new GPT-5.4 Cyber variant, analyzing the trend of restricted access for frontier capabilities. The episode also covers OpenAI’s Codex updates, including computer use and image generation features.
Topics discussed
Sponsors: OutShift and Notion AI Agent Platform
Intro, Host Updates, and Video Release Schedule
Anthropic Opus 4.7: Benchmarks, Deception, and Safety
Meta MuseSpark: Agent Architecture and Chain of Thought
AI Infrastructure: Power Demands and Deception Eval Concerns
Anthropic MIF: Cybersecurity and Access Restrictions
OpenAI Codex, Google Gemini, and Adobe Firefly Updates
Legal: Court Denies Anthropic's Motion to Block DOD Contract
XAI Co-founder Departs for New Infrastructure Startup
CoreWeave: Market Dominance, CapEx, and Data Center Delays
Cohere: Financial Struggles, Pivot, and New Pricing
Acqui-hires and Shoe Company Pivot to AI
Open Source: Tencent 3D World Models and NVIDIA Memory
AI Backlash: Attacks on Executives and Deepfake Misinformation
Research: Automated Weak-to-Strong Alignment Research
Research: Fragility of Vector Steering for Eval Awareness
Geopolitics: Iran Threats, Banking Vulnerabilities, and Outro
More from Last Week in AI
#261 - (belated) Argon, Muse Spark, Anthropic S-1
Nov 20, 2026
#260 - Muse, Misalignment, Math
Oct 19, 2026
#259 - Dots, Sonnet, Self-Safety, Rogue AI
Oct 9, 2026
#258 - Opus 5.5, Sol and Luna, Muse, DeepSeek-V4.1-Flash, Xi
Oct 2, 2026
#257 - GPT 6 Astra, AI Extinction, Security Incidents
Sep 24, 2026
Listen ad-free on Castria