Here's the abstract and summary for the provided transcript:
Abstract:
This video presents an exclusive first look and comprehensive comparison of the new ChatGPT 5 model against its predecessors, primarily ChatGPT 4o. The creator tests various capabilities, including raw response speed, complex coding (generating playable Tetris and Pokémon-themed chess games), factual accuracy (hallucinations), self-awareness, and creative tasks like YouTube thumbnail generation and scriptwriting. While GPT-5 demonstrates significant speed improvements and impressive "intuition" in coding, autonomously enhancing game complexity, it exhibits persistent hallucination issues in factual queries and shows mixed to poorer performance in certain creative and image generation tasks compared to older models. The video highlights OpenAI's new unified model approach, consolidating previous specialized versions into a single adaptive AI, and discusses the implications for both free and paid users. User comments reflect a divided opinion, with many praising its speed and coding prowess but criticizing a perceived loss of "personality," creativity, and increased hallucinations compared to ChatGPT 4o.
Testing ChatGPT 5: Speed, Intelligence, and User Experience
0:00 Introduction to ChatGPT 5: The video introduces ChatGPT 5, noting it will become the default model for all users (free and paid) and begins with a focus on its speed.
0:11 Speed Comparisons: Initial tests show ChatGPT 5 is significantly faster (30-50%) than previous models (e.g., GPT-4o) for simple factual queries, completing tasks in 3.1-8.3 seconds compared to 6.5-11.2 seconds for GPT-4o.
1:06 Unified Model Approach: The most significant change is that GPT-5 integrates all the capabilities of previously separate, specialized models (e.g., GPT-4o, O3, Mini) into one general model. It intelligently selects the appropriate underlying "thinking" model (light or heavy) based on the complexity of the user's query.
GPT-4o generated a basic Tetris game in 57 seconds.
GPT-4 Mini High (paid) generated a similar game in 14 seconds.
GPT-5 generated a much more complex and vastly better game (with score, level, upcoming piece, and controls) in 19 seconds, demonstrating "intuition" without explicit prompting for these features.
GPT-4 Mini High failed multiple times to create a playable Pokémon-themed chess game.
GPT-5, after explicit "thinking longer," successfully generated a fully functional game with themed rules, Pokémon evolutions, and legal move highlighting, showcasing a "leagues ahead" performance in complex coding challenges.
6:09 Persistent Hallucinations: Despite claims of improved factual accuracy, GPT-5 still hallucinates, inventing non-existent tech products from food brands (e.g., McDonald's XT mobile phone, Oreo smart speakers) with "complete confidence," indicating this area still needs significant work.
7:38 Improved Self-Identification: GPT-5 demonstrates better self-awareness, providing a clear and "human-oriented" answer when asked "What AI are you?", unlike older models that returned "gobbledegook."
GPT-5 performed worse than GPT-4o in generating a YouTube thumbnail, with poorer composition, less thematic text, and an incorrect aspect ratio.
Similarly, for a Star Wars themed birthday invitation, GPT-4o produced a "so good" and sophisticated design, while GPT-5's was "really bland" and showed "less effort."
9:49 Enhanced Writing Flare: GPT-5 shows significant improvement in writing style and delivery. When asked to create a "tech fail" script about the Windows Phone, GPT-5 provided a "much better starting point" with notes for filming and B-roll, using more engaging analogies and a less "wooden" tone compared to GPT-4o.
11:58 Overall Assessment & Access: While not universally better in every single aspect (e.g., image generation), GPT-5 is considered a "groundbreaking" launch due to its integrated capabilities and overall improvements. The monthly subscription price remains the same, and free users will get capped access (around 8-10 daily requests) to the most powerful version, then a "Mini" version.
Summary of YouTube Video Comments:
The YouTube comments section reveals a highly polarized reception to ChatGPT 5. Many users express significant disappointment, primarily lamenting a perceived loss of personality, creativity, and depth compared to ChatGPT 4o. They describe GPT-5 as "dumber," "less soul," "cold, flat," and "lobotomized," often missing the conversational style of its predecessor and reporting a failure to maintain context. Users frequently state they prioritize accuracy and detailed answers over speed, criticizing GPT-5's tendency to provide shorter, simpler responses.
Hallucinations remain a major concern, with multiple users reporting GPT-5 inventing information or providing incorrect details with confidence. Some users express discontent with the removal of the option to choose specific models (like 4o or Mini), preferring the control they previously had over which model to use for different tasks. This change has led some to consider cancelling their subscriptions or switching to alternative AI platforms.
Conversely, a substantial number of comments praise GPT-5 for its impressive coding capabilities, particularly its ability to generate complex, playable games like Tetris and Pokémon Chess from simple prompts. Some acknowledge its increased speed and appreciate its conciseness for certain tasks, while others commend the content creator for their transparent and direct review style. There's also curiosity and debate about whether GPT-5 was truly "released" or if the video represents early access, and some users note that image generation is typically handled by a separate AI model.