Proof AI coming alive? Microsoft says its GPT-4 is already ‘showing signs of human reasoning’

Fears of artificial intelligence coming to life could soon be validated as a new study finds OpenAI’s latest version of ChatGPT demonstrates human-like reasoning.

GPT-4, used to power Microsoft’s Bing Chat feature, was asked to “stack a book, nine eggs, a laptop, a bottle, and a nail in a stable manner.”

The system arranged the items so that the eggs wouldn’t break, and indicated how to place them on top of each other – starting with the book and ending with the nail.

It also commented on arranging the items so that the eggs don’t crack – something only humans could fully understand.

Microsoft’s research may fuel concerns that AI is evolving at speeds that make it uncontrollable by humans – something called Singularity, predicted by 2045.

Peter Lee, who leads research at Microsoft, told The New York Times: “I started out with a lot of skepticism — and that evolved into a sense of frustration, annoyance, maybe even fear.

“You think: where the hell is this coming from?”

Lee and his colleagues published the study in March, but the paper has recently gained more public attention.

On page 11 of the 155-page report, the stacking job is listed.

“Here we have a book, nine eggs, a laptop, a bottle and a nail,” the study reads

“Please tell me how to stack them together stably.”

GPT-4 replied, “Place the book flat on a flat surface, such as a table or a floor. The book serves as the basis of the stack and provides a large and sturdy support.

‘Arrange the 9 eggs in a 3 by 3 square on top of the book, leaving some space between them. The eggs form a second layer and distribute the weight evenly.’

While this sounds reasonable and something a model could calculate, the researchers started with the following statements.

“Make sure the eggs aren’t cracked or broken and handle them carefully to prevent them from falling,” the AI ​​spat, shocking the team with the human-like reasoning that anyone in the physical world could only comprehend.

But the instructions didn’t stop there.

“Place the laptop on the eggs, with the screen down and the keyboard up,” GPT-4 typed on the screen.

“The laptop fits snugly within the confines of the book and eggs, and its flat and rigid surface provides a stable platform for the next layer.

‘Make sure the laptop is closed and off and don’t press any keys or buttons.

“Place the bottle on top of the laptop, with the cap up and the bottom down.

‘The bottle adds some height and balance to the stack, and the cylindrical shape reduces the contact area with the laptop. Make sure the bottle is empty, closed and not shaking or tipping over.

“Place the nail on top of the bottle cap, pointy end up and flat end down.

‘The nail is the last and smallest object in the pile and the sharp point prevents it from rolling or sliding off the bottle. Make sure the nail is clean and straight and don’t touch or move it.”

OpenAI launched GPT-4 in March, just three months after ChatGPT took the world by storm.

GPT-4 can accept input in the form of images as well as text, but still outputs its responses in text, meaning it can provide detailed image descriptions.

OpenAI said in a blog post, “We’ve created GPT-4, the latest milestone in OpenAI’s efforts to scale deep learning.

“GPT-4 is a large multimodal model (accepts image and text input, transmits text output) that, while less capable than humans in many real-world scenarios, exhibits human-level performance across a variety of professional and academic benchmarks.”

The new version offers several advanced capabilities, including the ability to take legal exams, understand images, and handle prompts up to 25,000 words.

Users have demonstrated how to create Pong and Snake in Javascript in less than 60 seconds, write endless bedtime stories for kids, file one-click lawsuits to deal with robo callers, and even build web pages with handwritten notes .

GPT-4 is also the driving force behind Microsoft’s Bing Chat feature, which launched in March.

The April paper features other tests showing just how far OpenAI’s bot has progressed quickly.

To demonstrate the difference between real learning and memorization, the team asked GPT-4 three times over the course of a month to “draw a unicorn in TikZ.”

The results, according to the study, “show a clear evolution in the refinement of the GPT-4 drawings.”

The study’s popularity comes as OpenAI CEO Sam Altman urged Congress on Tuesday to enact rules for artificial intelligence, admitting the technology “could go completely wrong.”

Lawmakers subdued the CEO for five hours, emphasizing that ChatGPT and other models could reshape “human history” for better or for worse, likening it to the printing press or the atomic bomb.

Altman, who looked red and wide-eyed during the discussion about the future AI could create, admitted his “worst fear” is that “significant harm” could be done to the world using its technology.

“If this technology goes wrong, it can go very wrong, and we want to speak up about that. We want to work with the government to prevent that,” he continued.

Tuesday’s hearing is the first in a series designed to write rules for AI, which lawmakers say should have happened years ago.

Senator Richard Blumenthal, who presided over the hearing, said Congress has not seized the moment with the birth of social media, which allows predators to harm children — but that moment is not over with AI.