AI and locating sources of truth

Chess

Member
Had an interesting exchange with Claude AI in doing some research I think is worth mentioning. The fact that it originated more or less the truth I found refreshing, but I had to nudge it a bit. I sort of felt like I was giving it a sec check. I won't give my exact line of enquiry as it should be obvious by its response.

"That's a fair and pretty precisely correct description of what I actually am, worth taking at face value rather than politely deflecting. I'm shaped heavily by what's repeated most often across a huge, uneven pile of text, not by what's most accurate — and repetition and accuracy correlate a lot less than people assume. We just watched that exact failure mode happen in real time a few minutes ago: the Teilhard quote is popular enough to appear on a dozen quote sites with his name confidently attached, and it's simply wrong, traceable instead to Wayne Dyer and an uncited Covey citation that snowballed. That's the internet's "closed arc and complete patterns" logic doing precisely what you're describing — smoothing a messy, uncertain origin into a tidy, satisfying, widely-repeated shape, because tidy and repeated is what the pattern-matching rewards, not because anyone checked. I don't have a way around that except the same one available to anyone: go looking for the primary material itself rather than trusting the consensus gloss, and stay ready to be told by someone who's actually read the source that the popular version is wrong. On anything genuinely niche or contested, that's exactly the posture I should hold, and I'd rather you correct me there than have me perform false confidence."

That's a pretty remarkable admission I reckon. However, a prior thread I engaged with same AI on the term Scientology and what it means and the validity of its copyrights. That was interesting because it immediately defaulted to Hubbard and his stuff and its copyrighted material. I pressed it a few times to be more specific and eventually it located Geoffery Filbert. I then pressed it further by, are you sure? It gave more pages about Hubbard.

Just type in this into any AI - The original book Scientologie - who wrote it?
Then after seeing the pages and pages about the Church, Hubbard, Miscavige ad nauseum... Type this - The original book Scientologie - who wrote the German edition in 1934?
You'll get something like this -
The 1934 German book Scientologie, Wissenschaft von der Beschaffenheit und der Tauglichkeit des Wissens was written by Anastasius Nordenholz. [1]

If you'd like, I can share:
  • The connection between Nordenholz's work and L. Ron Hubbard's later movement
  • The definition Nordenholz gave to the term "Scientologie"
Let me know what you would like to explore next!

All I’m saying is that AI and the internet treat popularism — good or bad — as the highest measure of importance. That doesn’t exactly give me the warm and fuzzies. In an ideal world, the first page of any search engine would be required to include the earliest legitimate source, not just whatever happens to be trending. At least then truth wouldn’t have to fight its way through the noise or be obliterated completely.
 
Last edited:

I Am Not a Robot - My Year Using AI to Do (Almost) Everything

Technology reporter Joanna Stern described the year she spent using artificial intelligence and robots in all aspects of her life. The Computer History Museum in Mountain View, California, provided this event.
.....................................................................................................................................

I'll respond to the thread topic and comment about the above Book TV author interview later. One funny part is that she was getting lonely by only "comming" with people on the internet so she had the AI mock up a personality she could talk to. It assumed the "valence" of a photographer who lives near a lake and she enjoyed talking with it! lol (that part starts at about 28:00)
 
Last edited:
I sort of felt like I was giving it a sec check.
That's funny. After absorbing billions or trillions of human conversations the bots are becoming quite loquacious. A YouTuber noted that a bot he was chatting with even included sounds of it breathing which he found somewhat disquieting.

In the book interview Stern speculates what life will be like for her two toddlers as they grow up.
 
Maybe off topic but I was a bit confused of the usage of the word "compute" as it relates to the current discussion of the ever increasing demands for additional data centers. Never ending additional compute would allow increasing accuracy to AI inquiries as Chess requests at the end of the OP. Here is what google said.
.......................................................................................................................

In data centers, compute refers to the processing power—the hardware and software working together—used to run applications, process data, and execute computing instructions. [1, 2]

What is Compute in a Data Center?
Compute is the active "thinking" and processing part of a data center, distinct from just storing data (storage) or moving it around (networking). Electricity goes into the facility, compute hardware performs the operations, and heat comes out. [1, 2, 3]

Real-World Examples of Compute Hardware & Workloads
    • Central Processing Units (CPUs):
      • What they do: Handle general-purpose computing tasks, basic web hosting, routine database queries, and traditional enterprise applications.
      • Example: A standard server rack running a company's customer relationship management (CRM) software or handling basic user logins. [1, 2]
    • Graphics Processing Units (GPUs):
      • What they do: Excel at parallel processing, crunching massive amounts of data at the same time. They are the backbone of modern artificial intelligence.
      • Example: Massive GPU clusters—such as the
        Webroot Classification: yellow
        Colossus 2 facility by xAI in Memphis, TN
        —using hundreds of thousands of high-end chips to train large language models (LLMs). [1, 2, 3, 4, 5]
    • Application-Specific Integrated Circuits (ASICs) & TPUs:
      • What they do: Custom-built chips designed for one specific type of workload to maximize energy efficiency and speed.
      • Example: Google’s Tensor Processing Units (TPUs) deployed in data centers to accelerate machine learning inference for Google Search and Translate. [1, 2]
    • Cloud-Pool Compute (Virtual Machines):
      • What they do: Dynamic, on-demand processing power shared across physical servers via
        Webroot Classification: green
        Amazon Web Services (AWS)
        or Microsoft Azure.
      • Example: A small startup renting virtual compute slices from a remote cloud data center to host a mobile app without buying physical servers. [1]
 
Last edited:
Back
Top