[MagnusWootton]

locking ppl out of competitions is a devolution.

[LOCKSUIT]

Actually I might be wrong about the kind of extreme Turing Test I mentioned taking longer than Perplexity. I mean if you see some presented system can learn to ride a bike, research, solve problems, and conversate, for a few hours, isn't that enough? Why must one really test it on all tasks and each task 5,000 problems? Ex. summarize this, and that, etc, etc, etc...........translate this, and that etc, etc, etc..............etc etc etc etc etc? Surely, humans know an AGI by looking at it only for a while.

Maybe Perplexity only takes a few minutes or hours to run, while TT takes days/ weeks? And checking lifespan/ clone colony size takes too long for any of us to test? That last one takes little compute but long time to run. TT takes some compute/intelligence and some time - it's more middle ground, it is the best option? Perplexity takes lots of compute but is fastest. However Perplexity doesn't seem to take too much compute that it is infeasible to run at all, it seems like the best way to use most the time then. Any time you add a new part to your AGI, ex. embeds, or RL, you should be able to test it on thousands of tasks to see if it is scoring good.

[MagnusWootton]

Perplexity I just looked up the word,  is it how complex the data stream can be, before it becomes non predictable.

And its definitely semantically oriented, because u cant look at data redundancy in a general way,  its how you tackle the problem that makes it predictable, and that is data specific.

If u look at redundancy generally, u are guaranteed failure.   thats what I reckon.