News Google and Bing AI Bots Hallucinate AMD 9950X3D, Nvidia RTX 5090 Ti, Other Future Tech

If I were a big component vendor or movie producer, I would be annoyed at what the AIs are doing. Making up content about real corporations' products could lead to brand damage. I wonder how much longer until one of these AIs spreads NDA information from a litigation-happy company.
 
  • Like
Reactions: PEnns
So what you're saying it is acting like King Donny, just making things up to please it followers? It not so much Artificial Intelligence as it's an ABSA (Artificial Bull S**t Artist).
 
  • Like
Reactions: PEnns
I actually wonder how much of their future predictions might be from pulling BS speculation from Reddit. Plus people just typoing the product numbers.


But yeah the AI shouldn't be doing any of that. Logically it should be checking the Nvidia site first to see if the product even exists before even entertaining any information from other sources.
 
The AIs even offer specs, benchmarks and pricing for these imaginary products.

Google and Bing AI Bots Hallucinate AMD 9950X3D, Nvidia RTX 5090 Ti, Other Future Tech : Read more
I'd say they just work as designed. And that most surprises just stem from the fact that some people still don't understand how they are designed and operating.

These LLMs extrapolate on your input using data they were trained on. If their training data contained the relevant information and if that was actually correct in the first place, there is a good chance it will match what we believe to be the truth.

But it doesn't fall back into training mode to see if any more up-to-date information changes its prediction, because that would be far too expensive computationally, while it couldn't really keep fact and fiction separated either.

If currently sometimes it will correctly state that a certain iphone doesn't yet exist, that is very likely one of those hard coded overrides that the maintainers of these models are putting in to curb hallucinations or any other responses human audiences might consider aberrant.

But those moderator teams evidently aren't nearly as creative as the editors on this site, so they haven't nailed down overrides for every potential future product IT giants might deliver. I am pretty sure they generate data on iphone 42 internally, before the override takes hold.

I imagine these moderator overrides to be somewhat like classical expert system rules and there is a good chance that the noted quality loss in GPTx repsonses is actually due to these expert rules interfering badly with the quality of their extrapolation, in cases where that extrapolation was actually better.

People just shouldn't forget that AIs are artificial first and somewhat more intelligent than steam machines after. And intelligence literally just means that you fill the gaps, 'legere' means putting or laying down in Latin and "inter" means in-between.

Most of the time that is also just extrapolation, sometimes it's mindblowing abstractions and theories on general relativity. But don't expect the ability to develop such theories any time soon, I think it's a better match to think of current LLMs as the ability to extrapolate from all of Wikipedia (+Github, +some more) in real-time answers that are similar.

Factuality is hopefully a high correlation on the inputs the current AI "brains" have no capacity to do a "Gedankenexperiment" on their output before they utter it, as most humans would do beyond infancy.

Small children are much more likely to "hallucinate" in a similar way and we find that charming, sometimes even inspiring, while we work hard to override it with "common sense" and "fact checking".

AIs would have to gain that ability to operate in domains where today they clearly are out of their depth, but I'm also glad that the barriers to achieve that are probably unsurmountable for years to come.

And there is a good chance they will never be overcome, because AIs don't earn salaries, do not spend money or pay taxes. As humans get pushed out of the loop, progress comes to a standstill for lack of economic incentives.
 
  • Like
Reactions: Elusive Ruse
I actually wonder how much of their future predictions might be from pulling BS speculation from Reddit. Plus people just typoing the product numbers.


But yeah the AI shouldn't be doing any of that. Logically it should be checking the Nvidia site first to see if the product even exists before even entertaining any information from other sources.

I was thinking the same. The web is full fanboys who either bash or hope something from future products so it is rather easy to ”make news” about new products based on that fan-fiction that web is full…
There is less fan fiction about phones. So ”put bad data in, you get bad data out” still works like a charm!
 
Last edited:
The AIs even offer specs, benchmarks and pricing for these imaginary products.

Google and Bing AI Bots Hallucinate AMD 9950X3D, Nvidia RTX 5090 Ti, Other Future Tech : Read more
Again.. this just showdcases that "AI" is not even intelligent at all.
If you can poison their databases as easy as those kids in reddit did with the fake news bot.
And inventing these products in the future..

Yeah, I still laugh at those morons who are so enthusiastic their bots will replace support/help staff or news site writers.
 
And intelligence literally just means that you fill the gaps, 'legere' means putting or laying down in Latin and "inter" means in-between.
The MMLs, however, can still provide a more accurate etymological definition.

"The etymology of the word "intelligence" can be traced back to the Latin adverb "intus" meaning "inside" and the Latin verb "legere" meaning "to read, understand, gather ideas and information about someone or something." Therefore, intelligence is the ability to understand reality not in a superficial way, but going beyond, in depth, to capture its hidden and not immediately evident aspects. Another etymological interpretation (less widespread) of the term prefers the preposition "inter" = "between" to "intus". Therefore, intelligence would be the ability to read (...between the lines), to discover relationships and inter-connections between the various aspects of reality in order to reach a broader and more complete understanding of it." (not a MML generated text)
 
It's really just machine learning. ChatGPT actually extrapolated fairly well from a programming question one of my colleagues asked. Unfortunately, the functionality they requested didn't exist, so it just made up references pointing to where it should be.
 
"the bots answered as if those products were already on the market and had been benchmarked."

Sounds like lots of editors and members of many forums on the interwebz.....😛
 
Similar to when you'd ask a big-box electronics store for something that didn't exist, and get the "Hmm we don't have one in stock at the moment, but let me check with my manager and I'm sure we can order it in"....
 
The MMLs, however, can still provide a more accurate etymological definition.

"The etymology of the word "intelligence" can be traced back to the Latin adverb "intus" meaning "inside" and the Latin verb "legere" meaning "to read, understand, gather ideas and information about someone or something." Therefore, intelligence is the ability to understand reality not in a superficial way, but going beyond, in depth, to capture its hidden and not immediately evident aspects. Another etymological interpretation (less widespread) of the term prefers the preposition "inter" = "between" to "intus". Therefore, intelligence would be the ability to read (...between the lines), to discover relationships and inter-connections between the various aspects of reality in order to reach a broader and more complete understanding of it." (not a MML generated text)
So there you go and trust an MML more than me? Really...

And why do you think the word is interlegere and not intuslegere?

Intuslegere would definitely be more "in-a-box-thinking" than "inte[r]legere" and probably turn out as "introlegere".

All three variants exist, inter, intus, intro and perhaps various others. And they do have slight variations in meaning. But "inter" is really in betweeen, bridging, not already-inside (intus) or towards the inside (intro).

My first foreign language was Latin, I learned from Jesuits and for nine long years. It's not my strongest language by any means: contrary to the other four I don't speak it fluently.

That said, intelligence covers quite a long range. So people start using it once it can do the first thing level of intelligent behavior which is filling the gaps.

And I think we both agree that current AI isn't capable of abstractions yet nor of inventing something like a theory of relativity.