Skip to content

The future of ChatGPT

User: “What is the weather in Lima?”

ChatGPT: “I do not have access to real-time weather information. However, you can check the weather conditions for Lima by searching the Internet.”

LOOK: Can’t use ChatGPT? The four best alternatives to the OpenAI chatbot

This 2023 surprises us with the abilities of ChatGPT to strike up conversations about almost anything. And while this technology seemed to be reserved for sci-fi movies, last year there were already rambles about whether these algorithms are conscious. the future of ChatGPT it’s brilliant because it’s useful. And while there is a feeling that it could revolutionize the Internet, it is far from perfect and presents significant technical challenges ahead.

What is the future of ChatGPT? The magnitude of such a question requires understanding where this model stores knowledge. ChatGPT processes a set of documents the size of the Internet by interconnecting processing units called neurons, which encode the presence of words in a text. ChatGPT’s memory is distributed among the connections of its more than 175 billion neurons. Such a processing strategy makes algorithms appear smarter if the number of neurons, data, and processing are increased.

The problem is not that ChatGPT lack of memory but this is only long term. So while this algorithm learns to identify people, events, and dates in “[quién] won [la Copa del Mundial de fútbol] in it [2018]”, lacks a short-term memory to retrieve recent events such as the result of a soccer game, the value of the dollar, or the name of the new president. Today the solution to update your knowledge is to re-train ChatGPT, a process so expensive that the current version only contains information until September 2021.

In the future ChatGPT You will need to learn to connect to and search a rapidly changing stream of facts found in databases outside of the model. So while ChatGPT doesn’t know what the weather is in your city, Alexa and Siri respond correctly. Giving ChatGPT a short term memory will make it more personal as it will understand your location and preferences. Such a separation will also make it possible to clearly define the limits between public Internet data and private knowledge bases whose consistency is monitored by experts (weather, stock market, match results, etc.). This memory is more like the Knowledge Graph that currently feeds the searches of Google. It’s no surprise that Bing just released a hasty version of this concept, letting ChatGPT perform online searches and then decide when to trust your long-term memory or augment it with data coming from the Internet. The result is behavior with questionable interactions resulting in reduced search engine usage to 50 questions per day.

Having worked several years in implementing algorithms for AI and language models, I am surprised by the technological leap that represents ChatGPT. Until now, the commercial use of text-generative algorithms included machine translation and news summarization. And although ChatGPT has yet to solve other problems such as learning from text with images, or implementing reasoning and common sense, our society for the first time freely has an assistant that can fluently converse on all kinds of topics, even those for which it has not been trained. But for AI to benefit all of humanity, such enthusiasm needs to be accompanied by an understanding of its own limits, especially when these systems fail. The proliferation of web pages that misinform about medical and food issues are examples of the incorrect use of this technology. However, in the midst of so many technological news, let’s enjoy witnessing how the AI it makes it possible for us to converse with the knowledge that human beings have been accumulating on the Internet for so many years.

Source: Elcomercio

Share this article:
globalhappenings news.jpg
most popular