Most active commenters
  • ionwake(4)
  • dwattttt(4)

←back to thread

Nvidia won, we all lost

(blog.sebin-nyshkim.net)
977 points todsacerdoti | 12 comments | 04 Jul 25 21:58 UTC | HN request time: 0.208s | source | bottom
Show context
ionwake ◴[04 Jul 25 22:56 UTC] No.44468630[source]▶
I don’t want to jump on nvidia but I found it super weird when they clearly remote controlled a Disney bot onto the stage and claimed it was all using real time AI which was clearly impossible due to no latency and weirdly the bot verifying correct stage position in relation to the presenter. It was obviously the Disney bot just being controlled by someone off stage.

I found it super alarming because why would they fake something on stage to the extent of just lying.i know Steve jobs had backup phones but jsut claiming a robot is autonomous when it isn’t I just feel it was scammy.

It reminded me of when Tesla had remote controlled Optimus bots. I mean I think that’s awesome like super cool but clearly the users thought the robots were autonomous during that dinner party.

I have no idea why I seem to be the only person bothered by “stage lies” to this level. Tbh even the Tesla bots weren’t claimed to be autonomous so actually I should never have mentioned them but it explains the “not real” vibe.

Not meaning to disparage just explaining my perception as a European maybe it’s just me though!

EDIT > Im kinda suprised by the weak arguments in the replies, I love both companies, I am just offering POSITIVE feedback, that its important ( in my eyes ) to be careful not to pretend in certain specific ways or it makes the viewer question the foundation ( which we all know is SOLID and good ).

EDIT 2 >There actually is a good rebuttal in the replies, although apparently I have "reading comprehension skill deficiencies" its just my pov that they were insinuating the robot was aware of its surroundings, which is fair enough.

replies(7): >>44468687 #>>44468689 #>>44468696 #>>44468755 #>>44468868 #>>44469457 #>>44472494 #
elil17 ◴[04 Jul 25 23:05 UTC] No.44468687[source]▶
As I understand it the Disney bots do actually use AI in a novel way: https://la.disneyresearch.com/publication/design-and-control...

So there’s at least a bit more “there” there than the Tesla bots.

replies(1): >>44468725 #
1. ionwake ◴[04 Jul 25 23:11 UTC] No.44468725[source]▶
I believe its RL trained only.

See this snipet : "Operator Commands Are Merged: The control system blends expressive animation commands (e.g., wave, look left) with balance-maintaining RL motions"

I will print a full retraction if someone can confirm my gut feeling is correct

replies(3): >>44468845 #>>44469377 #>>44469719 #
2. dwattttt ◴[04 Jul 25 23:30 UTC] No.44468845[source]▶
Having worked on control systems a long time ago, that's a 'nothing' statement: the whole job of the control system is to keep the robot stable/ambulating, regardless of whatever disturbances occur. It's meant to reject the forces induced due to waving exactly as much as bumping into something unexpected.

It's easier to stabilise from an operator initiated wave, really; it knows it's happening before it does the wave, and would have a model of the forces it'll induce.

replies(1): >>44469018 #
3. ionwake ◴[05 Jul 25 00:00 UTC] No.44469018[source]▶
I tried to understand the point of your reply but Im not sure what your point was - I only seemed to glean "its easier to balance if the operator is moving it".

Please elaborate unless Im being thick.

EDIT > I upvoted your comment in any case as Im sure its helping

replies(2): >>44469046 #>>44469050 #
4. rcxdude ◴[05 Jul 25 00:06 UTC] No.44469046{3}[source]▶
'control system' in this case is not implying remote control, it's referring to the feedback system that adjust the actuators in response to the sensed information. If the motion is controlled automatically, then the control loop can in principle anticipate the motion in a way that it could not if it was remote controlled: i.e. the opposite, it's easier to control the motions (in terms of maintaining balance and avoiding overstressing the actuators) if the operator is not live puppeteering it.
replies(2): >>44469054 #>>44469084 #
5. dwattttt ◴[05 Jul 25 00:06 UTC] No.44469050{3}[source]▶
It's that there's nothing special about blending "operator initiated animation commands" with the RL balancing system. The balance system has to balance anyway; if there was no connection between an operator's wave command and balance, it would have exactly the same job to do.

At best the advantage of connecting those systems is that the operator command can inform the balance system, but there's nothing novel about that.

6. dwattttt ◴[05 Jul 25 00:07 UTC] No.44469054{4}[source]▶
Apologies, yes, "control system" is somewhat niche jargon. "Balance system" is probably more appropriate.
replies(2): >>44469101 #>>44473427 #
7. ionwake ◴[05 Jul 25 00:14 UTC] No.44469084{4}[source]▶
Thank you for the explanation
8. dboreham ◴[05 Jul 25 00:19 UTC] No.44469101{5}[source]▶
Well "control system" is a proper term understood by anyone with a decent STEM education since 150 years ago.
replies(1): >>44478259 #
9. elil17 ◴[05 Jul 25 01:20 UTC] No.44469377[source]▶
Only as opposed to what? VLAM/something else more trendy?
10. numpad0 ◴[05 Jul 25 02:38 UTC] No.44469719[source]▶
"RL is not AI" "Disney bots were remote controlled" are major AI hypebro delulu moment lol

Your understanding of AI and robotics are more cucumber than pear shaped. You're making very little technical sense here. Challenges and progress in robotics aren't where you think they are. It's all propagandish contents you're basing your understandings on.

If you're getting information from TikTok or YouTube Shorts style content, especially around Tesla bros - get the hell out of it at Ludicrous Speed. Or consume way more of it so thoroughly that you cannot be deceived anymore despite blatant lies everywhere. Then come back. They're all plain wrong and it's not good for you.

11. tekla ◴[05 Jul 25 15:29 UTC] No.44473427{5}[source]▶
> "control system" is somewhat niche jargon

Oh my god. What the hell is happening to STEM education? Control systems engineering is standard parlance. This is what Com Sci people are like?

12. dwattttt ◴[06 Jul 25 06:14 UTC] No.44478259{6}[source]▶
To be fair, lots of fields have a notion of a "control" system. Control Theory doesn't have a monopoly on the term, for all that the field revolves around 'control systems'.