The reason: Text generation applications such as ChatGPT or image generation applications such as MidJourney consume information faster than we can produce it. As a result, “high-quality” content that could be used to feed them will be discovered in 2026, according to a brief assessment produced in October by observers group from the artificial intelligence landscape.
In other words, the content that ChatGPT can produce may not cost much, but the content produced by humans is very expensive. The authors of this review take great care to highlight “qualitative” content, including books, scholarly research, and computer code. “It is economically costly to produce,” sums up one of these researchers, Pablo Villalobos.
Of course, there is also lower quality production, from YouTube videos to texts of all kinds circulating in the public space: the total amount of this other bit of information is difficult to measure, so that “the wall may not be reached” before 2050. But the designers of the artificial intelligence that It is relied upon to provide reliable information that they may be reluctant to leave shrouded in less content.
So far, these researchers wrote in their article Published in October on the pre-release ArXiv server, the data content used to feed this type of AI has grown by about 50% annually, while the so-called “quality” content available has only increased by 7% annually. Hence the wall of 2026.
However, in the language of these researchers, this AI is essentially an application that learns to speak, and so the quality of the “language data” may not matter as much as the amount of data used to train it. If so, Villalobos speculatesthe depletion of texts produced by humans can be compensated for by texts … by artificial intelligence.
“Music guru. Incurable web practitioner. Thinker. Lifelong zombie junkie. Tv buff. Typical organizer. Evil beer scholar.”