Thursday, August 7, 2025 – Launch Day
OpenAI unveils GPT-5 during a live presentation, showcasing its new “real-time router” that decides whether to respond quickly or spend more time thinking through answers.
During the event, a bar chart appears showing a lower benchmark score with a much taller bar, sparking immediate confusion and jokes online. The term “chart crime” begins trending among AI watchers.
CEO Sam Altman later posts on X, calling it a “mega chart screwup” and clarifying that the correct charts were in the official blog post.
Thursday Afternoon
Early users begin noticing GPT-5 responses feel less capable than GPT-4o. The culprit: a server issue disables the autoswitcher, meaning the model’s intended intelligence boost isn’t kicking in.
Friday Morning – Reddit AMA Begins
Sam Altman and members of the GPT-5 team host a Reddit ask-me-anything session on r/ChatGPT. Users flood the thread with questions about the rocky rollout, router problems, and the infamous chart.
Friday, Midday
Altman explains the technical glitch, confirming the router problem was fixed and promising tweaks to make the model-selection process more accurate. He pledges more transparency about which model is responding to each prompt.
Friday, Afternoon
Responding to repeated user requests, Altman says the team is “looking into” letting Plus subscribers continue to use GPT-4o. He also promises to double rate limits for Plus users as rollout completes.
Friday Evening – Closing Words
Altman ends the AMA with a pledge to “get things stable” and keep listening to feedback. Early reviewers like Simon Willison note GPT-5 is strong overall but still struggles with tasks like accurately formatting tables.
Aftermath
The “chart crime” becomes a meme in the tech community, with jokes about letting GPT itself make PowerPoint slides. For now, GPT-5 continues to evolve as OpenAI works to regain user confidence after its messy debut.
Leave a Reply