Google responds to ChatGPT with CALM, which promises to generate texts in less time

Model developed in collaboration with MIT cuts processing in half; The method tries to guess the next word with fewer guesses

OR Chat GPT stole the spotlight in recent weeks. The OpenAI chatbot uses GPT-3.5 artificial intelligence to give complete and articulated answers, which has caused some concern in the Google🇧🇷 The search giant took the opportunity to highlight its research in the sector: it is building the CALMa text generation method that promises to be faster and more efficient.

The model was called Confident Adaptive Language Modeling, or Confident Adaptive Language Modeling, in free translation. Hence the acronym CALM.

It is developed by Google and MIT and was presented at the NeurIPS conference, which took place in late November and early December.

Around the same time, OpenAI released its ChatGPT to the public. The potential of the tool has reportedly caused some discomfort for Google. There are those who think that this technology can in some cases replace the search engine.

Meanwhile, it is used for coding and flirting, among other things.

rescue processing

According to Google, leading linguistic AIs use the same amount of processing power on all word predictions, which can be thought of as killing mosquitoes with cannon fire.

“When writing a sentence, some continuations are trivial, while others require more effort,” writes the company on its blog. Then CALM saves when the next term is easier to hit.

Artificial intelligence works with the concept of jump start. Basically, when it finds a word that likely serves as a continuation of the sentence, the model calculates how much confidence it has in that word. If it is higher than the set limit, the choice is made and the algorithm moves on to the next word.

Be careful not to make a mistake

Google recognizes that some precautions need to be taken so that the AI ​​can write without knowing where it will go.

“Mistakes made at the beginning of the generation process are more harmful, because they affect subsequent choices”, reflects the company.

For this reason, CALM requires a higher confidence score at the beginning of sentences as a way to reduce risk. Subsequently, this cut becomes lower, as a way to move towards a better cost-benefit ratio, using the least possible energy for quality results.

Google claims that, in its tests, CALM has reduced processing times by almost half, without this meaning a loss of quality in the result.

For now the model is only an experiment and it is not possible to use it. Even so, it seems that Google didn’t want to let ChatGPT do the talking.

With information: Google search blog🇧🇷 Synchronized review🇧🇷

Google responds to ChatGPT with CALM, which promises to generate texts in less time

🇧🇷The best content in your email for free. Choose your favorite Terra newsletter. Click here!

Source: Terra

Camila

Camila Luna is a writer at Gossipify, where she covers the latest movies and television series. With a passion for all things entertainment, Camila brings her unique perspective to her writing and offers readers an inside look at the industry. Camila is a graduate from the University of California, Los Angeles (UCLA) with a degree in English and is also a avid movie watcher.