AI

GPT-5 is here. Now what?

While O1 was a major technological progress, GPT-5 is, above all, an accurate product. During a press briefing, Sam Germans compared the GPT-5 to the Apple retinal display, which is an appropriate analogy, although perhaps not the way he intended. It is very similar to an unprecedented screen, GPT-5 will provide a more fun and smooth user experience. This is not something, but it is no less than the future of transformational artificial intelligence that has spent most of them from last year. In briefing, Altman called GPT-5 “an important step along the way to AGI”, or artificial general intelligence, and it may be right-but if so, it is a very small step.

Take the demonstration of the form of the model showed by Openai Massachusetts Institute Technology Review Technology Before its release. Yann Dubois request, which is applied after training in Openai, GPT-5 Web app design that will help his partner learn freench so that you can communicate more easily with his family. The model did a great job in following its instructions and created an attractive and easy to use application. But when I gave GPT-4O almost an identical orientation, it has completely produced the same function. The only difference is that it was not aesthetically fun.

Some other improvements to the experience used are more fundamental. The existence of the model instead of the user chooses whether to apply thinking about each inquiry that removes the main pain point, especially for users who do not follow the LLM progress closely.

According to Altman, the GPT-5 is a much faster reasons than the Series Models O. The fact that Openai calls it to unpaid users indicates that it is less expensive for the company. This is a big problem: running strong price models at a low and difficult problem, and its solution is the key to reducing the environmental effect of Amnesty International.

Openai also took steps to relieve hallucinations, which were a continuous headache. Openai’s assessments indicate that GPT-5 models are less likely to submit incorrect claims than previous models, O3 and GPT-4O. If this progress communicates with scrutiny, this may help pave the way for more reliable and trustworthy agents. “Hells can cause real safety and security problems,” says Dawn Song, Professor of Computer Science at the University of California at Berkeley. For example, the agent who cheats software packages can download harmful software instructions on the user device.

GPT-5 has achieved the latest technologies on many criteria, including delegation capacity and Swe-Bus and Aider Polyglot. But according to Clemmanin Former, an artificial intelligence researcher in the company, these assessments are close to saturation, which means that current models have been achieved near the maximum performance.

“It is essentially similar to considering high school performance on middle grade problems,” she says. “If high schools fail, it tells you something, but if he succeeds, it does not tell you much.” Fourrier said it would be surprised if the system achieved 80 % or 85 % on the Swe-seat, but it managed to manage it by only 74.9 %.

Ultimately, the address from Openai is that GPT-5 feels better for use. “The feelings in this model are really good, and I think people will really feel that, especially the ordinary people who have not spent their time thinking about models,” said Nick Torley, president of Chattab.

Feelings alone, however, will not lead to the automatic future promised by Altman. Fear thinking as a big step forward on the way to AGI. We are still waiting for the next.

Don’t miss more hot News like this! AI/" target="_blank" rel="noopener">Click here to discover the latest in AI news!

2025-08-07 17:00:00

Related Articles

Back to top button