Advances in AI are coming at an incredible pace. Improved tools, new models, faster times, and we are entering the era of acceleration, however, amidst this euphoria, there is an almost imperceptible threat that looms in the shadows. What if it turns out that there’s a problem with the machines used for the experiments and production of AI? What if they fail?
According to this article The next phase of artificial intelligence may require very different processors, there are concerns that the next era of AI will not be based on a new model or a new tool but on a new processor. Honestly, it makes sense, today we use GPUs, which were not built for this purpose in the first place. We just adapted them to our use.
Like putting a sports car to the use of a moving truck, it works but until when? This is why several big companies and startups are already working to develop chips for AI, some are even emulating human neurons, the objective is for AI to be faster, more efficient and less energy-intensive.
The problem is that it takes a lot of computing power to train an AI model, in fact, it takes so much that if we continue like this, only the big tech companies will be able to afford the cost of training models. Not ideal, that’s why researchers are already warning that AI is going to depend a lot on computing power. The question of who has the most computing power is not a future problem.
Already today AI is reaching the limit of current processors, which is why big companies and governments are very aware that whoever has control of the hardware could have a huge competitive advantage. This is why this issue is already being discussed outside the field of technology. In short, the next big thing in AI could be the processor itself, not the model, it is no longer a matter of who has the best ideas, it is a matter of who has the most energy.