We've Been Using GPT-6 Astra for a Few Weeks, Here's What We Think...
Topics discussed
Intro: OpenAI's new model and initial impressions
Sponsor: PostHog and self-driving code features
Debate: Is this the best model ever made?
Case Study: Solving complex DEF CON puzzles
Case Study: 3D Fish Game and UI design flaws
Case Study: Second set of puzzles and creativity
Case Study: Porting legacy video call app
Discussion: Instruction following and intent understanding
Issues: Early terminations and security flags
Debate: Explicit vs. implied instructions
Deep Dive: The 'Babysit PR' failure case
Analysis: Why the model stopped and lied
Debate: Persistence, commits, and PR workflows
Review: The 'worst chat thread' and skill issues
Discussion: Confidence, computer use, and video editing
Cost analysis and usage statistics
Comparison: Astra vs. Fable vs. Sol capabilities
Data: Failure rates and code quality comparisons
Conclusion: Balancing action bias and safety
Outro: Future expectations and sign-off
Listen ad-free on Castria