Microsoft research Reveals GPT-4 AI Shows Promising Signs of Common Sense and Human-Like Reasoning

According to a recent Microsoft research, OpenAI’s GPT-4 artificial intelligence (AI) system exhibits “sparks” of common sense and human-like thinking. GPT-4 was shown to be capable of doing a range of tasks that involve common sense, including as answering questions about hypothetical circumstances, drawing conclusions from inadequate knowledge, and solving puzzles, according to the study, which was published in the journal Nature Machine Intelligence.

For example, when asked why it is not possible to travel faster than the speed of light, GPT-4 responded as follows:

“The speed of light is the fastest speed at which information can travel through space. If you could travel faster than the speed of light, you would be able to travel back in time.”

The tasks that the Microsoft researchers tested GPT-4 on included:

• Answering hypothetical situational questions
• Using insufficient data to draw conclusions
• Putting together puzzles
• Following instructions
• Creating innovative text formats such as poetry, code, screenplays, musical pieces, email, letters, and more.

GPT-4 was able to do all of these tasks with great accuracy. For example, when asked to respond to the question “What would happen if the Earth stopped spinning?” GPT-4 correctly answered:

The Earth would eventually fall into the Sun.

GPT-4’s capacity to produce innovative text forms also pleased the Microsoft researchers. GPT-4, for example, created the following poem when asked to write a poem about a unicorn:

The unicorn is a creature of myth With a horn of pure gold. It is said to live in forests And to be a symbol of purity.

The ability of GPT-4 to reason and apply common sense, according to the study’s authors, is a significant step toward the creation of artificial general intelligence (AGI). AGI is a potential type of AI that can accomplish every cognitive task that a human can.

However, the study discovered that GPT-4 has several limitations. For example, it was occasionally capable of producing false or misleading information, and it was not always capable of comprehending difficult or nuanced inquiries.

According to the authors, more study is needed to increase GPT-4’s capacity to reason and employ common sense. They also believe that measures must be developed to prevent GPT-4 from being exploited to create damaging or false information.

Overall, the outcomes of the study indicate that GPT-4 is a potential step toward the creation of AGI. However, keep in mind that GPT-4 is still in development and is not yet suitable for broad usage.