LLMs (AI in general)

There was a BBC Radio 4 programme on AI, specifically "model collapse". Something I read about in the New Scientist going back more maybe a year.

Basically "model collapse" is the AI analogue of positive feedback, or in an audio context "feedback" (which is technically positive - if not aurally). Positive feedback enhances, and with the complexity of an audio spectrum or the total content of the web - you can bet there will be places it quickly tips over into oscillation or polarisation (just as social media bubbles do)

The worst fact from the programme is that even the experts admit the percentages of enhancement (they call synthetic data!!!!) are rising rapidly. AI will be as honest as the web first and amplify its worst at best.

Will AI Et Itself My answer? - see above.

What I didn't pick up from the programme was that the language that the AI will preset with will devolve to Sun-reader level. (for the USA make that Trumpspeak)