On Sun, Mar 29, 2026 at 11:58:43AM +0200, Dr. Arne Babenhauserheide wrote:

[...]

> There’s a recent example about Schnapspralinen that made the rounds
> because it matched a prediction by a German Kinderbuchautor and
> Comedian.

[...]

Nice example. This shows how the lengthy training process introduces
a lag in the actuality of information (to be honest, the "Internet
at large" also has a lag -- probably everyone knows the experience
of looking for a local shop, finding references to it with address,
only to find out that they closed 11 years ago).

This is what Ihor was referring at with RAG [1], which looks like
a good technical idea to mitigate the lag issue.

My problem is rather with "AI with an agenda": being so eye-wateringly
capital intensive, only a few companies have access to the full
machinery. They'll do whatever is in their might to push their
agenda (and be it "just" to prove it is not a bubble, to generate
enough dependencies to get cash flow before *pop* happens; there
are other agendas out there).

Cheers

[1] https://en.wikipedia.org/wiki/Retrieval-augmented_generation

-- 
t

Attachment: signature.asc
Description: PGP signature

Reply via email to