Altman himself hinted at the arrival of Chat GPT-6 after the bizarre glitch with the goblins

Expectations for GPT-6 are focused on significant improvements, including more advanced memory, more consistent behavior, and deeper integration with different platforms and applications, which could change the way we interact with artificial intelligence.

The mystery of the goblins who haunted the code

It all started a few days ago when OpenAI found itself in a situation where it had to explain why its programming tool, Codex, was generating completely unexpected goblin-themed responses. The incident became a minor story in itself, and the company later adjusted the model to avoid such references. The solution was embedded in the very instruction (prompt) they gave to the artificial intelligence, which read:

“Never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other ž animals or creatures, unless absolutely and unequivocally relevant to the user’s query.”

OpenAI explained how this bizarre situation came to be. The problem arose during the training of the model for a specific feature of personality adaptation, specifically for the so-called “geek” personality.

“The short answer is that the model’s behavior is shaped by many small stimuli. In this case, one of those stimuli came from training the model for personality adaptation, specifically the nerd personality,” OpenAI explained in a blog post. “We unknowingly gave extra high rewards for metaphors with creatures. From there the goblins spread.”

Altman’s joke started an avalanche

Although the problem was solved, the story could have ended there if Altman himself had not joined the conversation. With his joking comment, he suggested that GPT-6 should get “extra goblins”. In no time, the blog about the technical issue with the goblins became secondary, and the Internet community started buzzing about GPT-6.

Although Altman did not explain in his commentary what exactly GPT-6 is or when it might appear, his speech was enough to confirm that the development of the next generation of language models is more important. in progress. This case with the goblins became memorable because it vividly illustrates how “human” these systems can still be, even while performing extremely technical tasks.

What does the new generation of AI models bring?

Altman’s mention of GPT-6 fits perfectly into the larger context of the development. It implies a version of the model that is not only more capable, butć is also more controlled and less prone to sudden digressions, such as an obsession with goblins. Apart from humor, moreć For some time, more serious hints about the future have been piling up. Altman has repeatedly described future models as systems with which we will have continuous interaction.

The idea is that the system remembers context, adapts over time and becomes a more permanent presence in our digital lives. Instead of opening the tool when we need it, we will live with it. That direction of development moreć is visible in features such as improved memory and deeper integration with files and applications that are implemented in newer models.

Utrka u naoružanju i kultura OpenAI-a

Another reason why Altman’s comment stands out is the timing. The time gap between the releases of the main models is significantly reduced. Competing companies like Google and Anthropic are indirectly pushing OpenAI forward, creating an intense race in the development of artificial intelligence. Talking about GPT-6 so soon after the release of GPT-5.5 seems almost inevitable in such an environment.

For users, this speed has an unusual effect – each new release becomes less and less exciting, but at the same time basic expectations are constantly rising. Features that once seemed revolutionary are quickly becoming standard. The informality of Altman’s announcement is also part of OpenAI’s culture. The gradual introduction of new ideas makes it easier for the community to absorb them. By the time the new model is officially presented, a large part of its contours will already be is visible in fragments.

The episode with the goblins fits into that pattern, showing both the limits and potential of the current models, as well as the bugs that still need to be corrected.

“Depending on who you ask, goblins are either cute or annoying model flies,” wrote OpenAI. “But they are also a powerful example of how reward signals can shape the behavior of models in unexpected ways, and how models can learn to generalize rewards in particular situations to those that are completely unrelated.”