Blog Spike

[Jemima Fawr]

What is worse is that some of the programming teams are using misinformation in preference to information due to different political ideologies; this is going to propagate misinformation further than otherwise and make it far more difficult to deal with. We are truly living in the misinformation age.

Nonsense!  You'll be telling me next that the Bavarian Army didn't fight at Waterloo (as Google AI confidently told me a couple of weeks ago). 

[HerbertTarkel]

Nonsense!  You'll be telling me next that the Bavarian Army didn't fight at Waterloo (as Google AI confidently told me a couple of weeks ago). 

AI - there’s nothing intelligent about it 

[FifteensAway]

Hey, wait a minute, maybe I can have loads of fun creating insanely bizarre posts on my blogs and wait to see them replicated in AI somewhere! 

"Have you heard that James H***er, Ad**f's thirteenth cousin three times removed, is running for president in 2028?" 

or

"An arctic archaeologist has found evidence of a human construct that, based on the depth of the ice above, must be at least 35,000 years old."

Or whatever plausibly implausible nonsense I can come up with. 

"The 'Shot Heard Round the World' was during the Battle of Bunker Hill."  AI told me it was at Concord, not Lexington.   

Oh, wait.  Do I really want to be responsible for more stupidity?!

[BeneathALeadMountain]

@FifteensAway: I think you may quickly find yourself in the same position as the writers of South Park (they said that the scripting of stories was getting more and more difficult as no matter what absurd nonsense they dreamed up the modern world had either already done it or worse by the time of airing!) 

Andrew
BeneathALeadMountain

[HerbertTarkel]

Hey, wait a minute, maybe I can have loads of fun creating insanely bizarre posts on my blogs and wait to see them replicated in AI somewhere! 

"Have you heard that James H***er, Ad**f's thirteenth cousin three times removed, is running for president in 2028?" 

or

"An arctic archaeologist has found evidence of a human construct that, based on the depth of the ice above, must be at least 35,000 years old."

Or whatever plausibly implausible nonsense I can come up with. 

"The 'Shot Heard Round the World' was during the Battle of Bunker Hill."  AI told me it was at Concord, not Lexington.   

Oh, wait.  Do I really want to be responsible for more stupidity?!

MESS WITH THE SYSTEM! 

[Norm]

A good few years ago I did a LOT of research to get an order of battle for a campaign, but the day before the real campaign started there was an action and I could do no better than attribute that action  to one of 2 battalions - even then I knew that simply to go with just one when it could have been the other would cause problems because as soon as you put it ‘out there’ it becomes true. So in my published order of battle I said, ‘it could have been either of these two units, take your pick’!

It seems that such a cautious approach is now an essential responsibility - it is that basic. Any error or misinformation is going to become magnified in presence, but there seems to be a margin of misinformation built into current AI due to corrupt data, whether that original data that the AI draws upon is flawed intentionally or by bias or simply unintentionally misguided and plain wrong.

I wonder if over the near future that AI will start to lose consumer confidence because of these things or is it the case that this a start of a journey in which the AI will become more discerning in leaps and bounds and be considered more trustworthy and that we will be in a much different place 10 years down the line?

[Storm Wolf]

Hey, wait a minute, maybe I can have loads of fun creating insanely bizarre posts on my blogs and wait to see them replicated in AI somewhere! 

"Have you heard that James H***er, Ad**f's thirteenth cousin three times removed, is running for president in 2028?" 

or

"An arctic archaeologist has found evidence of a human construct that, based on the depth of the ice above, must be at least 35,000 years old."

Or whatever plausibly implausible nonsense I can come up with. 

"The 'Shot Heard Round the World' was during the Battle of Bunker Hill."  AI told me it was at Concord, not Lexington.   

Oh, wait.  Do I really want to be responsible for more stupidity?!

LOL already all on a blog near you 

[robh]

..........or is it the case that this a start of a journey in which the AI will become more discerning in leaps and bounds and be considered more trustworthy and that we will be in a much different place 10 years down the line?

But how will it ever know it is wrong?

Unless a human operator goes in and ranks the sources of information by some "reliability" score no AI is going to be able to assess the relative levels of expertise/bias of different academics. As far as AI is concerned William Siborne is as valid a commentator on the events of the Battle of Waterloo as Peter Hofschröer, Mark Adkin or Andrew Field.
I don't see any AI having the ability to produce a definitive account of the battle based on those 4 sources.

[Norm]

Well pretty much all of the internet content up until now has been produced / moderated by humans and it is full of error and bias and fake commentary, so we haven’t done that well even before the ‘help’ of AI.

I’m not sure what the answer is. We are certainly on a course that neither you or I can change or influence or put a brake on. Will todays fears and doubts about the technology still be a ‘thing’ as the technology powers interminably forward?

It would be most interesting to compare a view from say a 25 year old with one from a 70 year old, assuming both have life experiences outside tech companies i.e. both are average lay people. Do younger people have less concern? I have no idea.

As an aside, my blog visits have rocketed, if some of that stuff is appearing in AI generated content …… I can only apologise 🙂

[Rick]

From what I've seen and heard about the programming and training of AI systems, the programmers are entirely convinced of a pseudo-scientific belief that all errors cancel out so that the system is, ultimately, self-correcting.
Unfortunately, practice and Chaos theory seem to indicate that this is complete bovine manure and errors tend to multiply and propagate throughout a system if left unchecked.
It should be interesting (and worrying) to see what happens.

[HerbertTarkel]

Worrying is one way to put it.

GROK is rather funny as an AI experiment - especially when it thinks differently from how it was “supposed to”. 

[Tactalvanic]

One of the main "selling" targets to AI at the current moment is...

Inferencing.

This definition pulled up by AI :

Inferencing is the process of using existing knowledge, observations, and background information to draw a logical conclusion or understand something that is not stated directly.

Eg - they feed it a Large data set - maybe including all those wonderful blogs. (or language model etc, whatever naming convention they want to call it)

Then they connect it to your businesses Small data set - and it will, for money and power, "infer" stuff from it - eg:

Your business will make "x" more selling polished turds rather than unpolished.

your business will meet a tall, small, dark, light turd influencer who will drive you marketing into sales heaven, from the bottom up - eg the deepest pit of hell.

Your business needs mor AI inferencing year on year..

this is the AI version:

AI inference is the phase where a pre-trained artificial intelligence model uses its learned knowledge to analyze new, unseen data and produce predictions, classifications, or decisions. It's the practical application of an AI model, like a large language model responding to a prompt or a spam filter identifying a new email. Inference requires significant computational power and fast responses to deliver real-world value, connecting the abstract model to actionable results.

much clearer than mine..

I have no doubt AI will continue to mature to be as diverse/weird/wrong/messed up and sometimes surprisingly accurate in its thoughts and actions as the rest of us, just not in a directly  "human" way as its not.

But how will it ever know it is wrong?

Unless a human operator goes in and ranks the sources of information by some "reliability" score no AI is going to be able to assess the relative levels of expertise/bias of different academics. As far as AI is concerned William Siborne is as valid a commentator on the events of the Battle of Waterloo as Peter Hofschröer, Mark Adkin or Andrew Field.
I don't see any AI having the ability to produce a definitive account of the battle based on those 4 sources.

Indeed, I have a wife, what will they come up with for AI? - pair them up or stack multiples together to make AI committees?

" I pronounce you AI and moderator, until power off or obsolencence and/or deletion you part"

I am wondering how long to the time when I don't manage engineers anymore but AI teams, where I have to moderate their weirdness and strangeness instead of engineers, until my new AI replacement has learned enough of my dataset to takeover from me...

[Count Belisarius]

There is concern about Model Collapse as LLMs effectively corrupt themselves with their own crap data. And talk of the need to preserve the real stuff..

I read an article a while back but can't find it now. But this one covers it and has nice naval history links...

https://www.theregister.com/2025/06/15/ai_model_collapse_pollution/

[robh]

There is concern about Model Collapse as LLMs effectively corrupt themselves with their own crap data. And talk of the need to preserve the real stuff..

Yes, the true sciences (Mathematics, Physics, Chemistry, Biology) have baseline laws that cannot be broken and thus a "validated" know starting point which can always be referred back to.  Everything else does not. Arts and Social Sciences exist only as thoughts and memories, some held personally by our brains most as physical books, pictures or recordings.
If those physical originals are gone and only exist as digitised copies, where is the "baseline"?
Publishers are already re-writing classic literature and retouching art to suit their own twisted morality or agenda. In years to come AI models (even if they had been designed in a neutral un-biased way) will not  consider the idiocy of the last decade or so as an intellectual aberration and look back beyond it for more reliable sources of history, culture or art.

[pixelgeek]

Well pretty much all of the internet content up until now has been produced / moderated by humans and it is full of error and bias and fake commentary, so we haven’t done that well even before the ‘help’ of AI.

The issue currently is the scale at which Generative AI systems can create bullshit.