Science and Tech

Google, Meta and OpenAI are running out of data to train their AI models, what alternatives exist?

Artificial intelligence

As if it were a human mind, artificial intelligences must be trained. That is the knowledge of this technology is “fed” using large data sets and machine learning algorithms. These may have different origins, but they all have something in common: that sooner or later they will run out.

That is not only an inevitable scenario, but one that the large technology companies that want to take the lead with AI, Google, Meta and OpenAI, view it with some concern.

However, far from standing still, they are already thinking about possible solutions to this problem. Some more surprising than others, it must be said.

How is AI actually trained with data?

Despite the controversies that have arisen as a result of alleged legal violations, the truth is that artificial intelligence is trained through various processes. But, without a doubt, the most important of them has to do with data acquisition, although then there is complex work so that the tool is able to validate and adjust them.

In any case, the first thing that companies do is to carry out a collection of large amounts of data relevant to the task that they want the artificial intelligence to perform. This data can be text, images, videos, transaction logs, sensor data, or almost any type of information you can think of that is useful in some way.

Taking this way of working into account, what will happen when there is no longer data to train artificial intelligence? Is it possible for something like this to happen? Actually, yes. This has been revealed by Epoch, a research firm that specializes precisely in everything related to artificial intelligence. And even has set a date for the event: it will happen in 2026.

So, if by then the AI ​​will have already “squeezed” all the existing information from which it can learn and train itself, what will happen from that moment on? The answer to that question could vary depending on the company you ask. Each of them already contemplates possible solutions in this regard.

New times, new data for artificial intelligence

Deposit Photos

The great technological powers are already thinking about where to get content from to continue feeding their insatiable artificial intelligences. In the case of Google, the plan to follow could be clear: pull his many tools, such as Google Docs, Google Sheets, Google Slides or Google Maps. Something that he already contemplated, although later he did not do it… at least for the moment.

The case of Meta is different, who seems to have remembered that knowledge is found in books. The responsible for Facebook or WhatsApp thought about buying Simon & Schuster, a major publisher that works with authors like Stephen King. They also do not rule out simply paying him to be able to directly access his work.

More complex is the option contemplated by OpenAI, the parents of ChatGPT. Sam Altman has already commented on some occasion that they could turn to synthetic information; That is, data that is generated by the artificial intelligence systems themselves. An extremely practical possibility, but one that would still be in development.

Thus, it is difficult to know how AI can continue to be trained in the future; what is clear is that, in one way or another, it will continue to be done. Provided, of course, that this technology does not cause the end of the world sooner, as some fear.

Source link