Want more intelligent visions of your inbox? Subscribe to our weekly newsletters to get what is concerned only for institutions AI, data and security leaders. Subscribe now
Associate Founder Openai and CEO Sam Al -Tamman publicly recognizes the main hiccups in Yesterday was affected by GPT-5The Pioneering Great Language Model (LLM) – It has been announced as stronger and capable so far.
Answer user questions On the theme of Reddit Ama (ask me anything) and In a post on x This afternoon, Altman admitted a set of problems that disrupted the launch of the GPT-5, including switching defective models, weak performance, and user confusion – Paying Openai to partially walk some changes on its platform and Returning the user to previous models such as GPT-4O.
“It was a little more rugged than we were hoping,” Al -Taman wrote in Responding to a question about reddit Regarding the Great GPT-5 launch.
as Wrong model performance plans It was shown during Openai’s GPT-5 Livestream, Altman He said: “People were working late and were very tired, and the adoption of human error. Many come together for live broadcast in the last hours. “
Artificial intelligence limits its limits
Power caps, high costs of the symbol, and inference delay are reshaped. Join our exclusive salon to discover how the big difference:
- Transforming energy into a strategic advantage
- Teaching effective reasoning for real productivity gains
- Opening the return on competitive investment with sustainable artificial intelligence systems
Securing your place to stay in the foreground: https://bit.ly/4mwngngo
While he pointed out that the publications of the accompanying blog and the system card were accurate, the mistakes that were launched after their launch are already facing scrutiny from the first users and developers.
Problems with the new automatic router
One of the main causes of the problem stems according to Altman from the new automatic router from Openai, which appoints the user’s claims to one of four GPT-5 variables, regular, mini, nano, and professionals-with optional “thinking” for heavier thinking tasks.
On xand Altman revealed that a major part of this system-Autoswitcher-was “outside the committee for a part of the day”, causing the appearance of GPT-5 “Way Dumber” from what was meant.
In response, Openai says it implements changes on the model decision -making boundaries and will make it more transparent any model that responds to a specific inquiry.
The user interface is also updated on the way to help users manually run the thinking mode.
In addition, Altman confirmed that Openai now will allow – After a wave of complaints about the inconsistent GPT-5 performance. he He said on Radit The company “Try to collect more data about the bares “before deciding the time it is presented to the old models.
yet Many users Incurred Beta Openai Test as Professor Warton College of Business Express the mixing and dismay in Unilateral Openai upgrade their ChatgPT experiments to GPT-5 and initially ignore access to old models.
Performance in the real world falls behind the noise
The internal standards of Openai GPT-5 may appear driving LLMS packageBut the real world users share a different experience.
Since launch, Users have published many examples of GPT-5 committing basic errors in mathematics, logic and coding missions.
Data scientist Colin Fraser has released the GPT-5 screenshots that are incorrectly solution if the repetition is 8,888 equals 9 (it is clear that it is clear)While another user showed that he extinguished the simple algebra problem: 5.9 = x + 5.11.
Other users have still reported a problem obtaining accurate answers to mathematics lyric problems or using GPT-5 to correct their presentation.
The developer’s notes were not much better, with Users publish pictures of GPT worse in the programming tasks “One shot” Complete them well with one obstacles- Compared to the new Claude OPUS 4.1 model of AI LAB artificial intelligence.
And the security company I found Splx GPT-5 still suffering from dangerous gaps To claim excessive logical injection and attacks, unless its safety layer is hardened.
Openai in the spotlight
with 700 million weekly users on ChatGPT, Openai is still the largest player in obstetric artificial intelligence by the audience.
But this scale brought increased pain. Note altman in his book x API traffic has multiplied more than 24 hours after the launch of the GPT-5, which contributed to the instability of the platform.
In response, Openai says it will limit a double rate for Chatgpt Plus, and continue to adjust the infrastructure because it collects comments.
But the early mistakes-which are doubled by UX’s changes and errors in confusion in the high-level launch-for them Open a competitor window to acquire the land.
Openai is pressed to demonstrate that GPT-5 is not just a gradual update, but a real step forward. Based on the initial start, many users are not convinced – yet.
https://venturebeat.com/wp-content/uploads/2025/08/ChatGPT-Image-Aug-8-2025-05_14_23-PM.png?w=1024?w=1200&strip=all
Source link