From Newsgroup: comp.misc

IrCOve been rereading old collections of rCLDilbertrCY cartoons, and then
this article <https://www.infoworld.com/article/4211958/anthropics-opus-language-problems-may-be-creating-a-hidden-cost-for-ai-coding.html>
popped up on my feed. Seems there may be issues with recent versions
of AnthropicrCOs Opus AI which cause it to, um, pad out its responses a
bit.

The overlap between truth and fiction/satire is getting downright
uncanny. From the first GitHub bug report cited:

* repeatedly propagates the terms "Load bearing", "prose"
(instead of text), "Hand-waving" (for being lazy/not putting
in effort), "Reflexive hedging", "Honest framing" , and lots
of other highly unusual terms it propagates as fact.
* Leads every sentence with what something isn't, instead of
what it is. E.g, "It is not Y. It is X."
* Made-up jargon / aphorisms presented as if standard:
"instrumentation is the unlock", "this is where a VP smells
hand-waving", "say the wrong expansion to a growth VP and it
dents you".
* Forced metaphors invented on the spot that require decoding
rather than aiding comprehension.

...

Real examples users quoted:

ruA "They're tightening, term-locking, and having the counter-probe
ruA answer loaded."
ruA "None of these are 'you don't get it' gaps."
ruA Fable 5: "The dice: clean rCo and one die never gets rolled
ruA anymore."

Heck, letrCOs face it: Scott Adams never came up with anything so far
outside the realms of normal human comprehensibility; it could very
well be that no normal human mind could go where the AI goes -- and
remain sane, anyway.

Looks like truth is leaving fiction in the dust, yet again ...
--- Synchronet 3.22a-Linux NewsLink 1.2