OpenAI, a US-based AI firm set up by Elon Musk and Sam Altman, released the preview version of a chatbot model called ChatGPT (Chat Generative Pre-Training) on November 30.
Immediately, the model took the internet world by storm, as the chatbot produced accurate, human-like answers (most of the time) to questions and commands.
OpenAI’s chief executive Sam Altman (left) and Microsoft’s chief executive Satya Nadella / Photo credit: Microsoft
In a nutshell, the model was trained using a form of supervised fine-tuning called reinforcement learning from human feedback (RLHF). That means human trainers provided conversational data for both sides – users and AI. ChatGPT could then learn from this exchange.
According to OpenAI, the latest release of the chatbot, which is now on version 3.5, is safer and more useful compared to previous deployments. For example, it can give more context on historical events, refuse to produce violent content, and give advice if questions from users have criminal intentions.
Currently, anyone can try ChatGPT for free, and people have quickly found creative uses for the tech, including writing music and writing code for building apps.
Users already say ChatGPT can replace the jobs of lawyers, programmers, and even poets. However, some argue that the tool is better used as a supplement to the work of humans.
See also: Series SEA: Who’s investing in the region’s AI startups?
But ultimately, the model has its limitations. For instance, the engine sometimes churns out plausible-sounding solutions that are incorrect – a product of it not being trained on factual information.
Stack Overflow, a question-and-answer platform for coders, has already set a temporary ban on ChatGPT-generated answers, saying that there’s been a massive influx of incorrect solutions from the AI tool.
There’s also potential ethics issues. During the tech’s ongoing testing phase, students have already been using it to write their essays for them.
“Overall, ChatGPT is a powerful language processing model with a number of strengths and limitations. Its ability to generate human-like text and learn from large amounts of data make it useful for a variety of applications, but its inability to provide accurate answers to factual questions and its reliance on large amounts of data limit its usefulness in some situations,” ChatGPT’s engine wrote when given a prompt to write an article outlining its strengths and weaknesses.
With its popularity, the chatbot likely won’t be free forever. In a reply to a question from Musk on monetization, Altman implied plans to charge users for each inquiry made.
Stay updated on the go with our mobile app.
Get latest insights with smoother, more personalized experience through TIA mobile app.
