While replying to an EV Semi truck post I consulted Google AI about the increase in total rig weight from 80 to 82 thousand pounds. It replied that I was wrong. I checked my question and noticed I had omitted to mention “EV” semi. Therefore the answer was correct but lacked intelligence enough to guess I was talking EVs.
I edited my question and it did know about the 82 thousand pounds EV allowance.
It’s early AI days!.
The Captain
No, nothing to do with early.
AI LLMs crystallize around the question of playing the odds. If you change the facts in the questions, the crystallization, as I call it, changes.
It can not answer two questions at once based on different wording because it never thinks. It tabulates the odds on every word.
We are all in deep crap when AI can guess our minds.
Ever watch the Hulu series “Paradise”? Cliff’s Notes: one scene people are talking about building and testing their new AI/Super Computer. First scientist says that the computer just solved some unsolvable physics/math problem. Second scientist says what’s wrong with that? First says we haven’t even asked the question yet.
I too checked Google and confirmed the 80000 lb (40 ton) limit on highways in most states. I find Google answers helpful. They do cite references. Experience is if you post an item on Wikipedia it will show up on Google the next day.
I see no indication Google answers include inferences. It merely translate your question into search terms and composites an answer—often as a one paragraph summary. But with extensive additional info available if you want it.
Wikipedia is asking for donations. Much of their info now goes to users via Google or Bing and customers are less aware it comes from Wikipedia and are less likely to donate. Wikipedia reports only 2% of users donate.
I agree but my point is that intelligent humans do “include inferences.” But for that to work the neural network needs to be larger or more of it accessible. Put another way, AI would look for other references to semis such as EV semis
These super-giant data-center neural networks are tiny compared to human brains.
The Captain
Google may be efficient in its improved searches. I don’t see much thinking involved. And no indication of fact creation. They do a pretty good job of scoring what you are looking for and presenting informative info.
If you want to see them at work ask them a controversial question such as was Nixon a good president. I think the response is balanced and not political.