Personally, I don't think the question is terribly relevant because I see no reason to believe AGI is possible with a binary logic playback device (aka a computer as we know it) and certainly not using relatively simple, statistical algorithms and lots of storage.
Expecting AGI to just spring forth from a box of inert electrical silicon-based switches is like expecting a hammer to start driving nails on it's own.
It's a modern form of alchemy --- expecting a fantastical result from trial and error with only limited understanding of the processes required.
That's the problem with intelligence. One day chess was considered a form of intelligence, that got solved by alpha-beta search and there was no longer a miracle.
Or is it that visual recognition models since YoLo are a miracle in the sense of practical image recognition?
>Prompt: Use your stateful python environment to count the letters in this comment: “Q: How many letters are in the word "intelligence"?
A: The word "intelligence" contains 11 letters: i-n-t-e-l-l-i-g-e-n-c-e.
Wrong! The correct answer is 12.
This is definitely not AGI! It is a crude statistical attempt that some people confuse with AGI.”
>Show your work. Then generate a Jupyter notebook that performs the calculations so I can double check, provide a link to download it from my storage when complete.
>Answer: The total number of letters in the provided comment is 187.
>You can download the generated Jupyter notebook to review and verify the calculations using the link below:
Just an aside, this was pretty impressive in my opinion.
I don’t know what AGI really means at this point, and I don’t think this is quite there, but this is an example of ChatGPT (4o) choosing the correct context (counting only alphabet characters not total character count), parsing the comment correctly (non-trivial given the usage of delimiters in both the prompt and comment and reference to counting letters in the comment not being interpreted as a part of the instructions), choosing the correct python tools, arriving at the correct answer, then generating and populating a working, downloadable, notebook to check the work. All in one reply.
It looks like you would need to continue the conversation from the link I provided then regenerate the prompt/reply to see the processing blocks and download the notebook.
Again, you need to define AGI. According to the classical definition of the term by the people who coined it, the example you gave proves that it is AGI. You were able to specify a problem using a fully generic interface (chat) that the AI was not specifically trained on, and it got an answer close to the right answer. The process it used to generate that answer is general intelligence.
This happens to be 11 tokens, but I think that's a coincidence. Token 491 is "int" and token 33465 is "elligence", but ChatGPT doesn't actually see the letters.
How can you expect it to count, given those limitations? It had to guess how many letters each token represented. It got close, but not exact.
This is an artificial example pretty much maximally designed for ChatGPT to screw up.
Expecting AGI to just spring forth from a box of inert electrical silicon-based switches is like expecting a hammer to start driving nails on it's own.
It's a modern form of alchemy --- expecting a fantastical result from trial and error with only limited understanding of the processes required.