What if AI had just reached a particularly perilous phase in its development? In any case, this is the opinion of Elon Musk and many other experts in the sector. Let’s take stock of this danger which could slow down the progress of this technology.
A major risk
During a live discussion with the president of the Stagwell company, Mark Penn, broadcast on X, this Wednesday, January 8, Elon Musk explained: “We have now exhausted virtually the cumulative sum of human knowledge… in training AI. This basically happened last year. »
The entrepreneur is particularly interested in this topic, since he himself launched the company xAI and his AI Grok is available on his own social network. So what can we do about this lack of data which could slow down the training and ultimately the progression of models?
According to the billionaire, synthetic data generated by AI is the solution: “The only way to supplement (public data) is to use synthetic data, where AI creates the latter. With synthetic data, it will somehow feed itself and go through this self-learning process. »
How to solve this problem?
This issue addressed by Elon Musk is a very widespread fear in the world of Tech. In fact, and according to data from Epoch AI, usable textual data should be exhausted by 2028.
Faced with this danger, specialized companies are turning to private data. It is in this sense that companies like OpenAI and Google have increased partnerships with media or certain sites dedicated to developers, such as Stack Overflow.
Finally, the idea of recovering synthetic data which would itself be generated by other AI is gaining ground. And in order to prevent errors and hallucinations from contaminating the process, human supervision will be more necessary than ever.
Moreover, several Tech giants, such as Meta and Microsoft, have already started using this type of content to train their models. According to Gartner, 60% of data has already been generated synthetically in 2024, which demonstrates that this movement is well and truly underway. To find out more, do not hesitate to consult our previous article on this subject here.
What to remember:
- According to Elon Musk, we have already exhausted the sum of human knowledge in the training of AI
- The entrepreneur believes that the solution to this problem is the use of synthetic data
- Several Tech giants are already massively using this process, which works provided it is supervised by human observers.
