Openai co -founder and The General Director of the Altman himself publicly recognizes the primary hiccups in yesterday’s implementation of the GPT-5The company’s recent flagship model (LLM) – advertised as its strongest and still capable.
Answering user questions In the Reddit AMA thread (ask me about anything) AND In the post on x This afternoon Altman confessed to a variety of problems that disturbed the launch of GPT-5, including defective models, poor performance and confusion of the user – prompted OpenAi to partial return of some of the platform changes and Restore user access to previous models, corresponding to GPT-4O.
“It was a bit more bumpy than we expected,” Altman wrote IN Answer the query about Reddit As for the great premiere of GPT-5.
As regards Incorrect model performance charts Shown during live broadcast GPT-5 OPENAI, Altman he said: “People worked late and were very tired, and the human error disturbed. In the last hours a lot has been connected live. “
AI scaling hits its limits
Power capitals, the growing costs of the token and inference delay are transforming AI Enterprise. Join our exclusive salon to discover how the best teams are:
- Changing energy into a strategic advantage
- Architect of effective inference regarding real capability profits
- Unlocking competitive roi using balanced AI systems
Secure your house to remain ahead: https://bit.ly/4mwgni
Although he noticed that accompanying the blog entry and the system card were accurate, errors moreover modified the premiere, which already had control from early users and programmers.
Problems with a recent automatic model router
One of the key reasons for problems according to Altman results from the recent automatic “router” OPENAI, which assigns the user to one of the 4 GPT-5-Regular variants, Mini, Nano and Pro-Z optional “thinking” mode for heavier tasks of reasoning.
On xIN Altman revealed that the key a part of this system-authorwitcher-was “outside the commission for a piece of day”, which meant that the GPT-5 seemed “stunning” than the intended one.
In response, OpenAI claims that he is implementing changes in the model limit model and will make it more transparent, which the model reacts to the query.
The user interface update is also on the way to help users manually liberate the way of pondering.
In addition, Altman confirmed this Openai will now allow ChatGPT Plus users to proceed using the GPT-4-year default model -after a wave of complaints about the inconsistent GPT-5 results. He he said on the reddit The company is “An attempt to collect more data about compromises “before deciding how to offer older models.
However, many users, including Openai beta testers corresponding to Wharton School of Business Professor Ethan Mollick Expressed confused and terrified Openai unilaterally modernizing your CHATGPT experience to GPT-5 and initially taking access to older models.
Real World Performance stays behind the noise
InternalBut real users share other impressions.
Since the launch, users have published many examples of GPT-5 committing basic errors in mathematical, logical and encoding tasks.
Data scientist Colin Fraser has published a screenshot of the GPT-5 Incorrect solution or repetition of 8,888 is 9 (in fact it is not)While one other user has shown that he is foubled a easy algebra problem: 5.9 = x + 5.11.
And other users have reported problems with obtaining accurate answers to problems with mathematical words or using GPT-5 to debug their very own presentation charts.
Feedback for programmers was not much higher, z Users publishing GPT images are advised worse in “one -time” programming tasks -Do it, complement them with one of them- Compared to the recent Rival Ai Lab Anthropic Claude Opus 4.1.
And a security company SPLX said that the GPT-5 is still suffering from serious gaps To obtain injection and darkened logical attacks, unless its safety layer is hardened.
Openai in the highlight
With 700 million weekly users on ChatgPT, Opeli stays the largest player in generative artificial intelligence by recipients.
But this scale brought growing pain. Altman noticed in his x post The API movement has doubled inside 24 hours after starting the GPT-5, contributing to the instability of the platform.
In response, OpenAI claims that double rate limits for CHATGPT Plus users and will proceed to adjust the infrastructure because it gathers feedback.
But early errors-prompted by misleading UX changes and errors in a loud launch-they He opened a window for rivals to get the ground.
Pressure pressure is included in OpenAI to prove that the GPT-5 is not only an incremental update, but a real step forward. Based on the initial implementation of many users, they are not yet convinced – yet.