Tech pundits now only talk about ChatGPT, which crossed 1 million free users within five days of its launch. ChatGPT is a conversational artificial intelligence based on a predictive language model called GPT-3, a model trained in 2020 on more than 500 billion texts from the Web, encyclopedias and books.
Behind this model is OpenAI, a research organization launched in 2015 to create artificial general intelligence, transformed into a profit-capped company in 2019, and which has been unleashing the passions of investors ever since.
ChatGPT’s ability to assemble, synthesize, prioritize all the content it has found on the Internet makes the craziest prophecies follow one another, announce the end of Google’s search engine, access to computer programming for all or the transformation of education. Especially since OpenAI is already communicating on GTP-4, its new model scheduled for 2023. And this company is just one of many alongside Cohere, GitHub, Hugging Face, DeepMind, not to mention Google, Microsoft, Meta , Tencent or IBM.
Alexa and Watson, the fallen stars
To judge the revolutionary nature of this tool and its competitors, the element that should really interest us is the growth in the number of testers, beyond the initial million concentrated on a circle of tech enthusiasts, while the application has been highlighted in all the media of the earth.
The number of daily users over time is also a factor, as other tools have grabbed public attention in recent years before disappearing. Think of Alexa or OK Google, the chatbots connected speakers that equip more than one in two American households, and whose use has disintegrated to the point that Amazon is laying off part of its team. It’s also a safe bet that you have already forgotten Watson, the IBM AI that won the general knowledge game Jeopardy in 2011.
An innovation thrives if it generates revenue. Some companies are already successfully using GPT-3 when they combine it with unreleased content. Created in 2021, Jasper, which produces marketing content using this model, forecasts $90 million in revenue in 2023 and just raised $125 million a few weeks ago on a valuation of $1.5 billion. of dollars. But the challenge of ChatGPT is to penetrate the world of work to become an easy-to-use productivity assistant. Only then will testers take out a monthly subscription capable of covering the costs of the computing power needed to run this tool.
A tsunami of mediocre content written by AIs
However, these concrete use cases remain to be tested. Of course, one person produced a children’s book using ChatGPT to write the texts from a few initial instructions, another had a few lines of computer code written by the chatbot, and another made him solve high school level mathematical equations, but value in our societies is created not by searching the Web, but by using hidden, confidential information, by making original and unpublished cross-checks. Yes, new content can be created from an individual’s original ideas that would be developed by ChatGPT, but the work provided by the creator is important.
More people could become creators, but the productivity gained is not enough for a marketing manager to become a successful children’s storyteller in his spare time. It will continue to be a full time occupation. As for the computer code, the quality of the code produced by ChatGPT, after the initial amazement, has been challenged to the point that the code-sharing platform Stack Overflow simply banishes it. Competitor DeepMind’s automatic code generation scored a mediocre 54% in Codeforces programming competitions.
More than the economic impact, it is the social impact that should concern us. Are we ready to live in a world overrun with abundant, low-quality content? Doesn’t this risk lowering trust between individuals, and even towards institutions within society? How can a teacher believe in the true talent of his students? I want to believe in the ability of human beings to filter information as its production increases.