There were times where I asked the same question twice and got slightly different answers, so yeah I think that's a thing.
You can also "corner" it, for example I was asking about a construction in japanese and there was an example that went slightly against what it had told me, so it insisted that the sentence was grammatically wrong. But then I closed the tab and asked about the same sentence again, without the context of the construction being talked about, and at that point it told me that the sentence is grammatically fine.
But like, if it had actually said that white people don't exist and black people exist, that would have been a little more weird than getting a slightly different answer in the way that you got Jankisa. I guess we would need an actual chud to really test that.
We can analyze, check locations, algorithms that existed before and how much weight they bring in for logged in users, user from different devices and locations, but in the end, even if we did all this very scientifically there is a high likelihood that we might get different results a few days later.
The fact is that in the end, we don't know how these machines "think" any better then how much we understand how we think.
We understand some of it's reasoning because it can be revealed, we can see, retroactively how different parts of the inference model get activated, similar to brain, but it's kind of a black box, in a lot of cases, we simply don't know and won't be able to know how and why these machines actually reach those conclusions.
The event horizon was back then when Lee Sedol got beaten by AlphaGo, we never got to understand how or why that very primitive (as compared to now) AI did some of the things it did, since then, we have been increasingly in the dark as to why these things do the things they do.
This Unitree ad is sorta like how RC car commercials used to be.
I also like the warning to not go near these things at the end lol
I feel like we need to have a global holiday/sickout/watchfest for all the ways this all can go horribly wrong and just try to take a deep breath and think about what we're doing as a species.
The video is pretty hilarious and obviously very staged, but also very impressive at the same time.
The "saving the kid from the car" bit where it just hits it with who knows how much force is crazy.
I think we are still very, very far away from these things being able to actually navigate the real world, but from what I've seen the LLM led AI revolution has already done a lot to improve the abilities of humanoid robots to navigate the real world.
On October 22 2026 00:34 Jankisa wrote: The video is pretty hilarious and obviously very staged, but also very impressive at the same time.
The "saving the kid from the car" bit where it just hits it with who knows how much force is crazy.
I think we are still very, very far away from these things being able to actually navigate the real world, but from what I've seen the LLM led AI revolution has already done a lot to improve the abilities of humanoid robots to navigate the real world.
Yeah, it very much reminds me of 90's toy commercials where they are very loose with their dramatic licence. But it also reminds me of the Will Smith eating spaghetti evolution.
Yeah, basically combining the agility of the Unitree bots with the learning capabilities of the cornhole bot isn't far away. Neither is the potential hyperscaling.
Feels like just a matter of time until we end up with a B1-66ER and I have no faith humanity will react well.
EPT
