| Mnementh said: Sorry that we germans make crazy streets. :) Routing is a *very* hard problem, I am impressed you worked on that. And you are right, before the current hype of LLMs all these different technologies were called AI. Currently there is a focus on neuronal networks and even more on transformer architecture. Which may be unhealthy for the field. I have a strong feeling the biggest progress will not come from advance in one area of AI, but from combining different AI technologies in a useful way. |
Oh Belgians are worse :p I came to an intersection there where all 4 directions had the same destination town name and Meisse was unreachable in the data. Due to turn restrictions you could only leave the town, not enter.
The human brain likely has different processes for different types of 'intelligence' as well. Neural networks are great for image processing, or sensory data processing in general. Transformer achitecture is great for root learning, learning a routine, reflex memory, which is basically what that Doom demo is.
Yet how higher reasoning, deduction etc work is still a problem.
And then how to tie it all together in an active loop which could be what we call consciousness. That voice in your head reasoning things out, combining inputs from sensory data, long term and short term memory to reason out a problem and come up with a plan / decision.
And we know all too well from humans that their decision making is based/biased on experience and beliefs. Hence the worry about using what data to train AI. Human history is fraught with bad reasoning and disastrous outcomes :/
Humans are also not a singular intelligence. Collectively we try to balance out our differences to arrive at a social contract and use each other's input to solve problems. Strength in diversity, same as evolution works. The danger with AI is, easy to clone, or rather one version. It's basically creating a dictator. Some dictators can be good, most aren't and certainly can't adjust well to changes.
Humans still haven't found a working system to live together in peace. How are we going to create a general intelligence that will...
Anyway LLMs can help with many things, good and bad. So far just another (very powerful) tool.
