> Most physical world thiigs have immediate feedback of incorrectness.
We might not want it to enter such circumstances where there is such feedback - e.g. car has run over a child. We generally aren’t interested in what the AI does in response to such feedback as such feedback should never ever occur.
Yeah 100% the initial announcement of Optimus was waaay to early, typical Elon bullshit to generate hype. But it is a real thing they are working on, and they have come a long way (considering they came from a guy in a spandex suit...)
I believe "Nvidia AI Lead" isn't accurate (and unnecessary here).
Author seems to have been working for only about 2 years since their PhD. While that shouldn't mean their opinion should be disregarded, it undermines the appeal to authority that is implied in the title.
16 comments
[ 3.5 ms ] story [ 59.6 ms ] threadWhat does this mean? Sounds like frantic bullshit
Most physical world thiigs have immediate feedback of incorrectness.
Bumping into something unexpectedly or falling over could be a built in way for models to self improve.
Self driving cars go WAY too fast for our current level of processing/sensors.
All IMHO.
We might not want it to enter such circumstances where there is such feedback - e.g. car has run over a child. We generally aren’t interested in what the AI does in response to such feedback as such feedback should never ever occur.
He lost me at this one. Tesla will need a lot more public verifiable proof of progress to look past the "guy in a spandex suit" thing...
Watch their latest update: https://youtu.be/D2vj0WcvH5c
Author seems to have been working for only about 2 years since their PhD. While that shouldn't mean their opinion should be disregarded, it undermines the appeal to authority that is implied in the title.