This is a ridiculous take. It is still doing anything a āstochastic parrotā would do. Being fed data about what towns are like and where supermarkets usually are, its simply waiting to recognise something like that and point you that way. It also didnāt ādirectā at the beginning.
Why are Metaculus markets now putting a 25% chance on AGI in the next 1.5 years?
Shit like this:
āHey GPT4-V, this is whatās in front of me. Letās step by step find the nearest supermarket using only logical reasoning.ā
Yes, thatās right - GPT4 was able to figure out how to direct a guy, step by step, to the nearest supermarket just by looking at a photo of a street, then using logical reasoning - piecing together clues from the tiny details in the roads, the signs, the cars.
āThis is the end of the road, should I go left or right?ā
This is why Geoffrey Hinton said in the next 5 years they may be better at reasoning than us, we will be the second smartest species on earth, and āthey might take over.ā
Anthropic CEO Dario Amodei thinks we're 2-3 years away from ~AGI. This is very roughly the same ballpark as OpenAI.
GoogleDeepMind's Chief AGI Scientist thinks 30% chance in 3 years.
We donāt have much time left.
If youāre still a stochastic parrot person at this point, what say ye about this example?