OpenAI is not happy with GPT-5 which could be delayed longer than expected

In recent years we have observed major and dizzying advances with regard to the artificial intelligencebut they are still not exactly infallible models.

That is why companies artificial intelligence They are working tirelessly on new versions of their well-known models, but according to Reuters It seems that they are not making the necessary progress.

In fact, Reuters comments that artificial intelligence companies like OpenAI are going to face delays, given that they are not finding adequate techniques to overcome their current models.

And now, some of the most prominent scientists in artificial intelligence point out the limitations when it comes to advances based on more and more data and more energy.

They comment that the main artificial intelligence laboratories have encountered delays. They say there is some disappointment because they are unable to release a large language model that surpasses current ones like GPT-4 of OpenAI which is almost two years old.

This is due to the issue of “training runs” for large models, which can end up costing tens of millions of dollars, and which are more likely to have hardware-induced failures given the complexity of the system.

Another problem that the main artificial intelligence laboratories are encountering is that these languages They absorb huge amounts of data and it seems that AI models have already exhausted all the easily accessible data in the world.

To solve this problem, researchers are exploring so-called “trial time computing,” a technique that improves current AI models during the so-called “inference phase.”

So instead of immediately choosing a single answer, the model can evaluate multiple possibilities in real timelike a person would do.

Interestingly, OpenAI has already opted for this technique with its recently launched o1 that can deal with problems in several steps in a way similar to human reasoning.

not only OpenAIbut also other artificial intelligence laboratories such as Anthropic, xAI and Google DeepMind are working to develop their own versions of this technique so that they can advance their current models.

Get to know how we work in ComputerToday.

Tags: Artificial intelligence