Uncategorized

Google has an artificial intelligence develop new AI chips

It’s amazing what has changed in computer technology over the past few decades. In the 20th century, computers filled entire rooms and could only carry out simple operations. Today we carry a powerful device the size of a small block in our trouser pockets. And we’re still a long way from the end.

Because so far the area of ​​a chip has limited its performance. After all, dozens of transistors and small data superhighways have to find their place within a few millimeters. After Apple, Google now also seems to be entering the business, with the next Pixel smartphone we could see a chip from the company in a device for the first time.

Perhaps more devices will soon have a chip from Google (Image: Google)

Previously, developing a new chip design took many months and consumed a lot of resources. Google now wants to change that and is taking an interesting path: Artificial intelligence is to develop the layout of future AI chips, and the new approach will be implemented for the first time with the next TPU (Tensor Processing Unit). Months of human labor could be completed in under six hours.

Researchers at Google trained artificial intelligence with various chip designs. The machine brain was fed both bad and good designs and developed its own suggestions from time to time. In addition to the space, waste heat and stability must also be taken into account.

Human (left) vs. machine (right) (Image: Mirhoseini, A. et al)

Initial results already show clear differences between the human approach and that of the machine. The images shown above have been made illegible for reasons of secrecy, but you can still guess how differently the computer thinks. In the future, Google hopes for faster innovations, the well-known Moore’s Law could be overridden.

Own opinion:

Computers design computers. For a long time that sounded like a vision of the future and sometimes brings dystopian thoughts. Nevertheless, it is clear that we can move forward even faster with ever more powerful machines. It remains to be seen whether the new approach will pay off or whether humans will outperform machines in the long term.

Via The Verge

Leave a Reply

Your email address will not be published. Required fields are marked *