In the last few weeks, a new technology called ChatGPT has caused great concern in the Google. It is an artificial intelligence to give complete and articulate answers. However, Google, not to be left out of this, developed together with MIT the CALM (Confident Adaptive Language Modeling) model.
Read more: Google Lens brings solution to doctors' historical problem: ugly handwriting
see more
Japanese company imposes time restriction and reaps benefits
Alert: THIS poisonous plant landed a young man in the hospital
If you are into technology systems, check out this release
It all started when Google realized that leading language artificial intelligences used the same amount of energy to process all word predictions.
In view of this, the company developed CALM (Confident Adaptive Language Modeling), which brings a technology able to save efforts when the next term is easier to hit during construction of a phrase.
This was a smart strategy, as the artificial intelligence works with the concept of early exit. If she thinks the next word is for use in a continuation of a sentence, the choice is made and then she moves on to the next one.
However, the technology company recognizes that it is necessary to take some precautions. This is because if the mistake is already made at the beginning of the word generation process, it is very likely that this will affect the following choices.
Therefore, CALM requires that there be a confidence score at the beginning of sentences, as this reduces possible risks. Afterwards, this requirement becomes lower as the construction of the sentence becomes more cost-effective, and thus the lower use of energy is also present.
According to Google, CALM has reduced processing time by almost half with no loss in quality. The model is not yet available for use, as it is in a test period.