GPT-5.6 just made itself better...

Matthew Berman explores the concept of recursive self-improvement by detailing how GPT-5. 6 Soul analyzed production traffic and identified inefficiencies.

He connects this to Andrej Karpathy's auto-research project and discusses the potential for AI labs to create a sustainable revenue cycle.

Matthew Berman maintains a clear and direct speaking style, using on-screen graphics to illustrate key data points and comparisons.

He explains complex technical concepts in an accessible manner, aiming to inform the audience about AI advancements.

Matthew Berman presents data and analysis from OpenAI and Artificial Analysis, while also acknowledging the potential implications of this technology, including his own nervousness about the pace of development.

He frames his own thinking as speculation at times.

Matthew Berman explains how OpenAI used GPT-5. 6 Soul to improve its own efficiency, leading to price drops for models like GPT-5. 6 Luna.

He details the implications of recursive self-improvement and the competitive pressure from open-source models like Kimmy K3.

Matthew Berman references specific data points from OpenAI's announcements and charts, including price reductions for GPT-5. 6 Luna and Terra, and cost per task metrics.

He also mentions a blog post detailing technical improvements.

Matthew Berman presents information using a standard talking head format with on-screen graphics to illustrate data points.

The visuals are clear and support the spoken content without significant distraction.

Matthew Berman concludes by emphasizing the power of recursive self-improvement and open source as the primary competitive pressure against leading AI labs.

He reiterates that the significant price drops for models like GPT-5. 6 Luna represent a major upgrade for users.