Brief Summary
This video reviews the new ChatGPT 5, highlighting its improvements over previous models. The key takeaways are:
- Speed and Efficiency: GPT5 is significantly faster for simple questions and intelligently selects the appropriate model for the complexity of the task.
- Unified Model: It combines the capabilities of various specialized models into one, streamlining the user experience.
- Enhanced Capabilities: GPT5 demonstrates improved coding abilities and creative problem-solving, though it still has issues with factual accuracy in certain areas.
- Accessibility: GPT5 is available to both free and paid users, with limitations for free users.
Speed Comparison
GPT5 is faster than the previous GPT4 model, especially for simple questions. In tests, GPT5 answered questions about Saturn's rings, the number of iPhones Apple has made, and Pokémon type combinations significantly faster, by 30 to 50%, than GPT4.
Unified Model and User Experience
The new GPT5 unifies various specialized models (like GPT40, 03, 03 Pro, O4 Mini, 04 Mini High) into a single general model. This eliminates the confusion of choosing the right model for each question. GPT5 can now determine the complexity of a question and use the appropriate processing power, optimizing speed and quality of response.
Coding Capabilities
GPT5's coding capabilities were tested by asking it to create a Tetris game playable in canvas. Compared to GPT40 and 04 Mini High, GPT5 produced a more complex and enhanced version of the game with features like scoring, level display, and upcoming piece previews, demonstrating its ability to deliver better results without specific prompting. When asked to create a playable game of chess with Pokémon sprites, GPT5 outperformed 04 Mini High by generating a functional game with highlighted moves and turn indicators, showcasing a significant improvement in coding and creative problem-solving.
Hallucination and Factual Accuracy
The issue of AI "hallucination," where chatbots confidently present incorrect information, is addressed. Despite improvements in hardware, user feedback integration, and benchmarking, GPT5 still struggles with factual accuracy. When asked to list tech products made by food brands, it invented products with complete confidence, similar to previous models. However, GPT5 showed improvement in understanding the user's intent when asked "What AI are you?", providing a more human-oriented answer compared to the older AI.
Creative Tasks and Image Generation
GPT5's performance in creative tasks, particularly image generation, was evaluated. When asked to create a YouTube thumbnail for a Mr. Who's the Boss video, GPT40 produced better results than GPT5. Similarly, when asked to create Star Wars-themed 30th birthday invitations, GPT40's design was more sophisticated and visually appealing. These tests indicate that GPT5 is not necessarily a step up in image generation compared to its predecessor.
Writing Flare and Script Generation
OpenAI claims that GPT5 should have improved writing capabilities. When tasked with writing a Mr. Who's the Boss tech fail about the Windows phone, GPT5 generated a script with notes for filming and B-roll shots, which was a significant improvement over GPT40. The writing in GPT5 used analogies effectively and provided a better starting point for a script, indicating enhanced writing flare.
Overall Conclusion and Accessibility
GPT5 is a notable upgrade with improvements in speed, efficiency, and coding capabilities. It integrates the complexity of multiple models into one simple interface. GPT5 is available to both free and paid users, with free users having a limited number of requests per day using the most powerful version, after which they will use a less powerful GPT5 Mini. The monthly subscription price remains the same.
