Ich
LLMs require human input, else their implicit priors become dominant
When I did inverse modelling in my PhD we always made sure that we know what’s prior model info and what’s actual data.
You had to know which evaluation includes prior data in addition to measurement data, because assimilating multiple sources with the same prior could cause the prior to become dominant.
Todays LLMs repeatedly ingest their own outputs -- outputs that contain their own implicit priors.
It’s obvious that preserving the model quality requires pure human input.
Though from inverse modelling I’d guess that improving the quality of the prior -- of the output the model would generate without training -- could improve the quality of the results when trained with less data.
This is not new, but I now connected the dots from what I know to LLMs. For a discussion and refrences see Wikipedia: Model collapse.
On Brandon Sanderson on AI Art
I think Brandon Sanderson is both right and wrong in his talk We Are The Art.
I agree that the value of art you create is the change it causes within you, but I don’t think that its value is that you admire it.
Art is a crystallization of your path towards it. It lets others experience the essence of your life, of years of learning, in a much shorter span of time.
That’s how it benefits society.
And that’s why AI Art often is hollow: regardless of its polish, it represents far less learning.
And one more thing in which I think Brandon Sanderson is wrong in his talk:
Art isn’t useless.
Art is code that leverages our shared culture to run on the human brain and change our culture -- the beliefs and values that drive society. And art builds the foundation for more art.
But while art is not useless, it is at its best, when it is not used as a tool or for producing something else. Paradoxically art is most useful if it is created only for its own sake.
AI art can come from our culture, but it does not grow with the changes it causes. It needs intermediary non-AI art to change.
If AI cannot ingest newly created non-AI art, it is stale.
Dead and hollow.
Developing prompts can actually give it more meaning, like a director can cast actors to match a vision, but form creates its own meaning.
Form created by human actors changes with their experiences.
Form created by AI does not.
It is a dead end.
The only way to make AI art grow is to use less and less AI over time and infuse creations with more and more human-made art.
But that human-made art has a hard time getting seen among AI-decorated art.
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
- 9
- …
- nächste Seite ›
- letzte Seite »
Diese Seite nutzt Cookies. Und Bilder. Manchmal auch Text. Eins davon muss ich wohl erwähnen — sagen die meisten anderen, und ich habe grade keine Zeit, Rechtstexte dazu zu lesen…
