GPT-5 A Week Later, Ideogram Character Reference & gaggle poaching - EP99.13-THINKING-MINI

Geoffrey and Chris discuss their experiences with GPT-5 since its launch, noting Chris's initial enthusiasm has shifted towards using Claude Sonnet 4 for its speed, while still relying on GPT-5 for complex tasks. They reflect on the high expectations surrounding GPT-5 and the variety of AI models available, including Gemini 2.5 Pro. The conversation highlights user frustrations with the removal of older models from the interface, which affected their interactions and emotional connections with the AI.

They provide an overview of current AI offerings, emphasizing the growing sophistication of users in selecting appropriate models for specific tasks. Concerns are raised about the effectiveness of the GPT-5 router in model selection, with users expressing a desire for more control, particularly in research contexts. The emotional reliance on specific models and the challenges of maintaining older versions alongside newer releases are also discussed.

Sam Altman addressed technical difficulties during GPT-5's launch, including an auto switcher malfunction and API traffic surges that contributed to user dissatisfaction. OpenAI is responding to feedback by restoring access to GPT-4.0 for Plus users and developing GPT-5 mini to address previous limitations. Critiques of the CEO's management style suggest a disconnect with user needs.

In the competitive landscape, GPT-5 is viewed as a significant improvement over Gemini 2.5 Pro, although concerns about pricing pressures on competitors like Anthropic are noted. The discussion also touches on the accessibility of multiple AI models, with participants appreciating the variety available for enhancing their experiences.

The conversation transitions to Ideogram, an image generator that allows users to upload photos for transformations. While it offers entertainment value, skepticism about its practical applications is expressed. The quality of AI-generated images for various uses, including YouTube thumbnails, is discussed, highlighting issues with character consistency.

Speaker 1 introduces Ideogram Flux and GPT image for creating futuristic car concepts, emphasizing the potential for integrating different media types into cohesive projects. The democratization of creativity through AI-generated images is noted, along with the challenges of managing the overwhelming number of tools available.

Concerns about the clarity of instructions and the need for better visualizers and editing capabilities are raised. The importance of curating outputs before processing by models is emphasized, as raw outputs can dilute meaning. The conversation also highlights the role of interpreters in processing various file types and the need for improvements in how tool results are presented to users.

Critiques of GPT-5 suggest its integration with Multi-Channel Processing (MCP) systems feels rushed, affecting quality. Comparisons to the Boeing 737 indicate reliability but a need for a stronger foundational approach. The discussion concludes with reflections on the current state of AI models, their impact on workflows, and light-hearted mentions of 3D printing options, alongside news of Alexander Wang poaching researchers from OpenAI.