TBPN GPT 4.5 Comparison, Where's the Moat?, RIP Skype, Sulek Earns Pro Card, Cursor for Figma
8 snips
Feb 28, 2025 Dive into a playful yet insightful discussion about the latest in AI, as GPT-4.5 takes center stage amid humorous critiques. The fall of Skype is nostalgically revisited, highlighting communication's evolution. A journey into the world of bodybuilding showcases personal dedication that mirrors tech innovation. Excitement builds around new creative tools like Flora, while discussions on the ethics and competition in AI provoke thought. Plus, enjoy lighthearted tales of travel adventures and the quirks of modern banking!
AI Snips
Chapters
Transcript
Episode notes
Coogan's Eval
- John Coogan evaluated LLMs by asking for funny jokes.
- The resulting jokes, while not written by John, were deemed not funny by Jordy Lee.
Subtle GPT 4.5 Improvements
- GPT 4.5's improvements over GPT 4 are subtle, similar to GPT 4's improvement over 3.5.
- The improvements are more qualitative (e.g., better word choice, nuance, analogies) than drastically improved performance.
Public vs. Expert Preferences
- Public preference for GPT 4 over 4.5 in some creative tasks contradicts Andrej Karpathy's expert assessment.
- This difference highlights a potential gap between perceived quality and expert evaluation in AI.