15 Discoveries of Early Signs of Artificial General Intelligence (AGI) in GPT-4

This article has been buried and I wanted you to see latest research from Microsoft. Researchers have found recently that GPT-4 is being viewed as a possible early form of (AGI) Artificial General Intelligence (Sparks of Artificial General Intelligence: Early experiments with GPT-4)

Here are the main 15 points of this article that were discovered by these researchers:

  1. Tool Use: GPT-4 is able to use tools like a calculator effectively with minimal assistance and produce an accurate answer.
  2. Text to Image Comprehension: Compared to its predecessor, GPT-4 can create a detailed image when asked and can understand how to arrange the objects in the image based on the given text.
  3. Coding: Successfully passing mock technical interviews, GPT-4’s ability to code is comparable to human performance. If human, it could be hired on as a software engineer.
  4. 3D Game Development: ChatGPT is unable to produce any type of digital game whereas GPT-4 can produce a game in a zero-shot fashion.
  5. Mathematics Whiz: GPT-4 successfully solved extremely difficult math questions from the 2022 Mathematics Olympiad. Problems it was unable to completely solve were ones that required more creativity.
  6. Fermi Questions: Out-of-the-box questions that have no easy answers such as how many golf balls fill a pool are easily answered by GPT-4.
  7. Personal Assistant: GPT-4 is able to manage one’s calendars, emails, coordinate with others and their schedules to book a meeting, lunch at a restaurant, and more.
  8. AI ‘Handyman:’ With the ability to perform “diagnostics,” you can ask GPT-4 how to fix/troubleshoot things like a leaky faucet.
  9. Mapping: GPT-4 can create a map of the place you’re entering such as a house layout with enough information and questions.
  10. Theory of Mind: It is about understanding beliefs, GPT-4 is able to separate what is true and what a human believes to be true. This allows it to build a mental picture of what someone is actually thinking.
  11. Why its Jokes Don’t Hit the Mark: GPT-4 has trouble with directions that require discontinuous steps, it can easily complete an incremental task, but it does not have the ability to think ahead and plan its output.
  12. Misinformation: GPT-4 can create entire false scenarios complete with images and links.
  13. Data Admission: People are asking for the ability and right to decide which GPT-4 content they do or do not want to be crawled.
  14. Intrinsic Motives: Not motivated anything, GPT-4 is not passive. This can potentially create both ethical and safety concerns if not taken carefully.
  15. Urgency: The urgency behind AI are the unknowns. Researchers do not really know what is happening and why AIs like GPT-4 are able to do what they do.

Top executives in the digital technology realm, like Elon Musk and Apply co-founder Steve Wozniak, are calling for a pause on AI development and research until we have a better understanding of AI itself and its potential dangers and risks if not handled properly.


Sparks of Artificial General Intelligence: Early Experiments With GPT-4. 23 Mar. 2023, https://www.microsoft.com/en-us/research/publication/sparks-of-artificial-general-intelligence-early-experiments-with-gpt-4/.