After a summer of chaos, Openai returns

Openai was only its best week in months. He was in urgent need of him.

The San Francisco -based company, based in San Francisco, which is famous for Chatgpt, has spent most of June and July in the main headlines of all wrong reasons. The Talent Raid came: Mark Zuckerberg, CEO of Meta, opened the check book, according to what was provided by hundreds of millions of dollars as compensation to attract senior researchers in Openai. Several ships jumped. CEO Sam, publicly criticized, describing him as a mercenary dead approach and accusing him of not having a culture.

Then came the failed acquisition of Windsurf, a incendiary startup specialized in infrastructure of the original data of artificial intelligence, which was in Openai in the purchase talks. Google has passed at the last minute and the deal was closed instead, a humiliating loss in the high AIMS race.

On his head, Openai had to delay the launch of its open source models long ago after the intense pressure of the developers, which has caused criticism that the company was leaving its competitors such as Meta, which issued its own models for free.

Internally, things seemed messy. The leadership handed over all employees a week, described the notes that were leaked by a company under siege, or a fortress attacked from all sides, or worse than that, a burning house. Dear artificial intelligence, which was applied once, began to appear, and imagined that Mita had stolen her momentum was growing.

From panic to axis

This week, Openai has finally started playing the attack again. First, it has released the long -awaited open source models, a step aimed at appearing in the developers and reaffirming its importance in the open AI ecosystem. Just three days later, the biggest swing came: GPT-5 launch, which was described as the most powerful Chatbot Amnesty International in the market.

Openai claims that GPT-5 treats two of the largest complaints about artificial intelligence assistants: “hallucinations”-when Chatbots spit with confidence in wrong information-a very nice tag that makes them look like companies in companies. The company says that the new model is faster, more accurate and able to provide more accurate answers without sugar. By learning to say “I don’t know”, GPT-5 aims to be the first Chatbot Amnesty International that you can really trust.

Although independent tests are needed to confirm these allegations, Openai’s operating program gave something that he had not had for weeks: controlling the narration. Currently, AI Spotlight returned to San Francisco, not in Menlo Park, where Meta is called the so -called “DREAM team” for former researchers building its own models.

At the same time, the company is discussing with investors about the sale of huge employee shares that would estimate it at $ 500 billion, a step seen widely as a defensive strategy to create “gold handcuffs” and stop the exit of talent.

The big question: Was this just a good week, or the beginning of a real return? In the world of high -speed artificial intelligence, stability rarely continues for a long time.

While Openai’s ambitious claims about GPT-5 still need to verify them, this week this week was unambiguous: while its competitors were writing checks and overfishing talent, Openai was building. With these two main releases, the company has effectively regained control of the conversation.

The lighthouse of artificial intelligence still illuminates the brightest in San Francisco, not in Menlo Park, where there is a “dream team” in a dead woman of artificial intelligence mercenaries. The question now is whether this strong symptom of strength is sufficient to end deviations and permanently restore momentum.