Microsoft Proves GPT-4 Can Think Like a Human
Experiments with GPT-4 have shown that the model can start the development of general AI.
IN Microsoft conducted a study in which they talked about how GPT-4 learns to think like a person. Report on 155 pages contains an analysis of how the language model copes with various tasks at the human level. The researchers also tested how well GPT-4 understands human psychology and logic, and conducted a series of experiments with interesting queries.
For example, the researchers asked GPT-4 to explain how to properly put together a book, some eggs, a laptop, a bottle, and a nail. In response, the AI model gave instructions that took into account physical laws and common sense.
According to Microsoft, this response shows that GPT-4 isn’t just learning, it’s taking it to the next level. The researchers wondered if this was a step towards the advent of artificial general intelligence.
“We believe that GPT-4 can be considered as an early (but not yet sufficient) version of the general AI system, given its wide and deep capabilities,” the experts write.
Modern systems are capable of performing one or more similar related tasks. It is assumed that general AI will be able to solve problems that are understandable to the human mind. The study has so far been published only on the arXiv preprint site and has not been peer-reviewed.
In March, more than a thousand famous personalities and specialists, including Elon Musk and Steve Wozniak, signed an open letter , calling on all AI labs to suspend training on more powerful AI systems than GPT-4. In the letter, a number of AI researchers claim that AI systems “with intelligence rivaling that of humans” could pose a serious danger to society and humanity.