GPT-5 failed to test noise

Last week, on the day of the GPT-5 launch, the noise of artificial intelligence was at its highest level ever.

In a journalistic briefing, the CEO of Openai, Sam German, said that the GPT-5 is “something I do not want to return from”, a milestone closer to The first iPhone with a retina screen. The night before the announcement of the live broadcast, altman to publish A picture of the star of death, building more noise. On x, one user books The expectation “looks like Christmas.” All eyes were on the ChatGPT maker, where people waited across industries to see if advertising would provide or disappoint. And on most accounts, the large detection will shorten.

The noise has been built for the old OPNAI model for years-since the 2023 version of GPT-4. In Reddit Ama with Altman and employees last October, users have constantly asked about the date of the GPT-5 version, looking for details about its features and what will dismantle it. “Why does GPT-5 take a long time?” Altman replied that Compute was restrictions, and that “all these models have become very complicated and we cannot charge many things in parallel as we wish.”

But when the GPT-5 appeared in Chatgpt, users were not greatly affected. It seems that the great developments that they were gradually expected gradually, and the main gains of the model were in areas such as cost and speed. However, in the long run, this may be a strong financial bet for Openai – albeit less glamorous.

People expect the world of GPT-5. (X user one to publish After every person died of the death star, “everyone transformed expectations.”) Openai did not reduce these expectations. GPT-5 call “The best artificial intelligence system to date” and “a big leap in intelligence” with “newer performance, mathematics, writing, health, visual perception, and more.” Altman said in a press conference that chatting with the model “seems to talk to an expert at a doctorate.”

This noise was made for a blatant contradiction with reality. Will it be held with a PhD level, for example, Insist over and over again There were three “bs” in the word berries, as some social media users found? Is it? Don’t be able to get to know How many names of states included the message “R”? Is it Incorrectly naming An American map with makeup states including “New Jefst”, “Micann”, “New Nakamia”, “Krizona” and “Miroinia”, and Nevada’s knowledge as an extension of California? People who used the robot for emotional support found that the new system is volatile and far, as they protested so highly that Openai brought support for an older model. Many memes – one imagine GPT-4 and GPT-4O as two huge eloquences with GPT-5 next to them as Simpleton.

The public opinion court was not forgiving expert. Gary Marcus, a pioneering voice in the industrial intelligence industry and professor of honorary psychology at New York University, The form is called “I was late, and surpassed the overwhelming.” Peter Wildford, co -founder of the Institute for Politics and Strategy of the Institute of Artificial Intelligence, books In his review, “Is this the huge destruction we were looking for? Unfortunately, no.” Zvi Mowshowitz, famous artificial intelligence blog, And invite her “A good model, but not great,”. Redditor on the official GPT-5 Reddit Ama books“Someone tells Sam 5 is hot garbage.”

On the days after the GPT-5 version, the attack of the unaccounted reviews slightly reduce. The general consensus is that although the GPT-5 was not important to progress as people expect, it provided promotions in cost and speed, as well as a fewer hallucinations, and the automatic automatic switch system on inquiries on the form that made the most logical to answer, so you did not decide-completely. Altman bent in this narration, writing“GPT-5 is the smartest model we have ever done, but the main thing we paid for is the benefit of the real world and the ease of access to the mass/ability to afford costs.”

Openaii Kristina Kim Researcher to publish On x with GPT-5, “The real story is the benefit. It helps in what people-the code of shipping, creative writing, and the movement of health information-with more stability and friction less. We also cut hallucinations. It is better to calibrate,“ I don’t know, ”separates the facts from guesses, and we can want answers with the procedures you want.”

There is a widespread understanding, to put it frankly, make the GPT-5 eloquent. Viral social media publications complained that the new model lacks the differences and depths in its deletion, as it runs out of the mechanism and the cold. Even in GPT-5 marketing materials, the comparison between the GPT-4O and GPT-5 generated from GPT-5 does not seem an uncommon victory for the new model-I prefer personally from 4O. When to be safe Request Redditors If they thought that GPT-5 was better in writing, it was met with an attack from the comments that defend the retired GPT-4o model instead; Within one day, he was pressed and repeated at least temporarily to ChatGPT.

But there is one interface where the model appears to shine brighter: coding. One repetition of GPT-5 Top currently The most popular AI in the coding category, with human Clauds in second place. Openai’s launch promotion showed games created from artificial intelligence (a small, circulating ball game, writing speed), pixel art tool, drum simulation, and Lofi. When I tried to deal with a puzzle game with the tool, she had a set of defects, but I found success with simpler projects such as interactive embroidery lesson.

This is a big victory for Openai, as it is directly heading in artificial intelligence coding wars with competitors such as Anthropor, Google and others for a long time now. Companies are ready to spend a lot on coding artificial intelligence, and this is one of the most realistic revenue generators for emerging companies to burn money.

Openai also highlighted the GPT-5 in the field of health care, but this is often not testing in practice-it is likely that we do not know how successful it is for a while.

Artificial intelligence standards have become less and less in recent years, because they often change and some cherry-bic Cherry companies. But in general, they may give us a reasonable picture of the GPT-5. The performance of the model was better than its predecessors in many industry tests, but this improvement was nothing to write about, according to many people in this field. As wildford Put it“When it comes to official evaluations, it seems that the GPT-5 was largely expected-the increasing small increases instead of anything worthy of the mysterious death star.”

But if modern history has anything to say on this topic, then these increasing small increases may be more likely to translate into tangible profit more than dissolved individual consumers. Artificial intelligence companies know that the biggest ways of making money are customers of institutions, government contracts, investments, additional payments forward on strong standards, in addition to investing in coding and hallucinogenic control, is the best way to get more three.