GPT-5 is really coming this time.
Now, the entire internet is "encountering" GPT-5.
Not only is there a trace on ChatGPT:
Moreover, it was also found in the model list of MacOS applications.
On Cursor, Microsoft Copilot, and OpenAI API platforms, netizens have also caught that GPT-5 has been quietly integrated/tested.
This wave of leaks has once again brought the excitement and curiosity about GPT-5 to a high point. Altman is really playing it well.
Even Markus, who always challenges everything, says: No joke, GPT-5 is really coming this time.
(The translation continues in the same manner for the rest of the text)Although GPT-5 will far exceed GPT-4 in capabilities, it will still be difficult to control and will continue to make unpredictable low-level errors, and may not even fully follow instructions.
GPT-5 will still struggle with some complex physical, psychological, and mathematical reasoning tasks.
Hallucinations will remain widespread and more misleading, making it easier to generate outputs that appear correct but are actually incorrect.
Natural language output will still be unable to reliably link to downstream systems (such as databases, virtual assistants, etc.), and natural language instructions will be difficult to accurately map to user intent.
GPT-5 will not be a general artificial intelligence like AGI, and will still require other tools to assist in completing complex scenario tasks.
The system will still be unable to consistently follow beneficial and harmless human values, and may implicitly contain some biases in its output responses.
GPT-5 is just a basic scale expansion, and future AGI will develop towards more structured, explicit knowledge and planning capabilities, integrating with a broader range of other technologies, which are lacking in the GPT series.
Similar views are also pervasive in the community, after all, Altman has slipped up too many times, and if it's not released soon, netizens are getting tired...
Every day is filled with mixed true and false leaks, and some leaked GPT-5 benchmark tests are fake news. Netizens say that now, apart from the official source, no one can be trusted.
Of course, many netizens also believe that these "unintentional" leaks are just OpenAI's hype.
In any case, regardless of whether the model is as good as advertised, just release it already!
Reference links:
[1]https://www.geeky-gadgets.com/openai-gpt-5-lobster-ai/
[2]https://garymarcus.substack.com/p/what-to-expect-when-youre-expecting-62e
[3]https://x.com/nicksomed/status/1950507062091485551[4]https://www.bgr.com/1926255/gpt-5-leaks-reveal-more-features-ahead-of-launch/?utm_source=chatgpt.com
This article is from the WeChat public account "Quantum Bit", author: Focusing on Cutting-Edge Technology, published with authorization from 36kr.
