[WriterOfMinds]

https://thegradient.pub/an-epidemic-of-ai-misinformation/

In Rebooting AI, Ernie Davis and I made six recommendations, each geared towards how readers - and journalists – and researchers might equally assess each new result that they achieve, asking the same set of questions in a limit section in the discussion of their papers:

Stripping away the rhetoric, what does the AI system actually do? Does a “reading system” really read?
How general is the result? (Could a driving system that works in Phoenix work as well in Mumbai? Would a Rubik’s cube system work in opening bottles? How much retraining would be required?)
Is there a demo where interested readers can probe for themselves?
If AI system is allegedly better than humans, then which humans, and how much better? (A comparison is low wage workers with little incentive to do well may not truly probe the limits of human ability)
How far does succeeding at the particular task actually take us toward building genuine AI?
How robust is the system? Could it work just as well with other data sets, without massive amounts of retraining? AlphaGo works fine on a 19x19 board, but would need to be retrained to play on a rectangular board; the lack of transfer is telling.

[ivan.moony]

Although I find neural networks alone insufficient to create AGI, I believe it could be an important piece of the puzzle which still misses better control over synthetic decision making.

[Art]

I found the part where two experts felt that GP2 was not equal to anything close to AGI but that it might be better suited for Troll farming or joke writing. Nothing really of substance.
Nice article.

[goaty]

gpt-2 will carry out things approximately due to its inefficient method of "knowing"   if you ask it whats the answer to a mathematical question,  itll start jabbering out numbers instead of doing the mathematics for real, but I bet that form of approximate reaction could still be useful in simple situations of life where it accidentally has all the bases covered in its text chains (and generalizing methods), with its humoungous garbage load of memory required.

NN's could still work for the real thing, just someone needs to work out something better than random trial and error.   My best advice would be what a lot of other ppl are already doing,  just working instead of just procrastinated about.

There is some fixed intuitive thing that is the fundamental of all things,   and then this is what is put into relation with itself,   Ive got the excuse that im working on something easier and need to get it done, like open ai, but maybe this proper way of learning doesn't take ten million gigs of computation, and everyone is barking up the wrong tree,   but maybe its good we are, because we need to get it out of the way before we move onto the more important way of doing it,  for example the computer actually generating an idea for real.