AI that learns without human input? Scientists make a STUNNING discovery

It is the era of artificial intelligence. Chances are high that you may have used text and image generation tools such as ChatGPT and DALL-E, solely because of the chatter they have generated.

Scientists studying these tools, however, are perplexed that the AI language models powering them can learn and answer without having been trained previously.

Add WION as a Preferred Source

Large Language Models like OpenAi's GPT-3 powering ChatGPT and Google's LaMDA powering Bard are exceptionally good at learning and executing tasks on the fly.

Meaning, unlike the linear process of training an AI machine by feeding it copious amounts of data which is a laborious and time-consuming process, the new breed of AI can dig deeper into its neural schema and arrive at outputs which it hasn't been specifically trained to perform.

Read more: Google Bard vs ChatGPT: Which is better AI tool? Know how to access Google chatbot, all you need to know

This apparent phenomenon is called 'in-context' learning and researchers from Massachusetts Institute of Technology, Stanford University, and Google in a recent study have set upon to decode how AI tools seemingly work between the input and output layers.

“Learning is entangled with [existing] knowledge. We show that it is possible for these models to learn from examples on the fly without any parameter update we apply to the model," Ekin Akyürek, the lead author of the study was quoted as saying by Motherboard.

Researchers said that the LLM is building upon its previous knowledge, just the way humans do. In fact, the models build smaller models inside themselves to achieve new tasks, posited the scientists.

WATCH |Race for powerful AI tools heats up

To prove their hypothesis, the researchers analysed a transformer, a neural network model. After observing the model in action, they asserted that the transformer was writing its own machine-learning model in hidden states. Long story short, the transformer was inventing the wheel, all by itself.

Read more: $100 bn wiped off Alphabet's shares after Google AI chatbot 'Bard' flubs answer

Machines learning on their own and adapting on the go is a wonderful discovery, one that enhances the limit of what the AI tools can do. However, experts are also wary that leaving information to automated systems brings its own sets of challenges.

There is still a shroud of mystery regarding what are the true capabilities of AI learning models but the discovery of 'in-context' learning has certainly widened the horizon.

(With inputs from agencies)

WATCH WION LIVE HERE: