As you have probably already heard, within the field of artificial intelligence there is a concept that is equally attractive and scary: Artificial General Intelligence (AGI).
Simply put, it refers to a point in the future when machines could perform almost any task that humans currently do.
Logically, it sounds a little scary if you really think about it, but calm down, because According to research led by Yann LeCun, the “chief AI scientist” at Meta, it seems that we are still far from seeing machines reach human intelligence.
As he explains and despite the many voices that are already beginning to mention this concept now that that mysterious OpenAI Q* project is being seen, current artificial intelligence, even with the existence of advanced language models such as GPT-4, is still can’t come quite far from matching these unique abilities.
Don’t be confused: this study shows that AI is far from reaching human capabilities
The study, co-written by researchers from various AI companies, such as Hugging Face and AutoGPT, wanted to carry out an evaluation as an exam to see where this technology really is today.
Simple questions were designed for humans but initially complex for advanced AIs, related to aspects such as reasoning, multimodal management and web navigation. The results, although not peer-reviewed, demystify the belief in widespread AI in the short term.
Research shows that, in these situations, humans outperform AIs by 92% compared to the 30% success of the most advanced AIs. They explain that humans generally outperformed large language models when it came to these more complicated real-world problem-solving scenarios.
“We hypothesize that the advent of Artificial General Intelligence (AGI) depends on a system’s ability to exhibit robustness similar to that of the average human on such issues,” the study concludes.
