Delays in the development of GPT-5: OpenAI faces challenges related to computing power

In the 1990s, early experiments with neural networks ran up against the limitations imposed by the algorithms and hardware capabilities of the time. A first breakthrough came in 2000. Today, however, the situation is very different: the level of technology achieved is already remarkable. We may not reachAGI (General Artificial Intelligence), but current AIs are already being used to make hardware more efficient.

Despite this, significant limitations still remain. Recently, Sam Altman of OpenAI announced that the launch of GPT-5 will be delayed precisely because of the lack of sufficient computing power. The increasing complexity of these models requires enormous computational resources, and OpenAI plans to collaborate with Broadcom to develop a proprietary AI chip, planned for 2026, that will enable it to support advanced models. In the meantime, improvements in the functionality of thecurrent technology are still expected by the end of the year.

Current development is not limited to the technology behind ChatGPT: another promising area is biocomputers, which in simple terms means the use of "test tube brains." Our brains, in fact, consume only 20 watts of energy, but can handle enormous processing and storage capacity. Unlike computers, the biological brain does not separate memory and processing and has a three-dimensional structure in its connections, a feature that contributes to its unique efficiency and ability to adapt and compute.

Recently, a Swiss startup connected 16 human brain organoids to create a low-power biocomputer, demonstratingnature's efficiency over traditional bit-based computers(Science Alert). In addition, scientists have developed a hybrid biocomputer called Brainoware, which combines lab-grown human brain tissue with conventional electronic circuits, capable of performing tasks such as speech recognition(Nature).

It's too bad I'm not younger: as a student I was passionate about expert systems and early neural networks; today I can only follow developments from practical, social and philosophical perspectives.