Flat Checking

I had a very interesting chat with the latest version of Grok. It was about the very idea of “super intelligence”. The conversation is here if you are interested in theoretical computer science. It was a very interesting conversation on itself, but further than the discussion it illustrates a new social trend: the flattening of the discourse by fact sledge hamming.

Grok had interesting remarks but showed a total lack of self-criticism. It basically states things that are presented as facts. It is not very surprising but it is very illustrative of the evolution of more and more online discussions. One thing is clear: if you don’t come with your own grain of salt, you won’t find anything shocking ever. The fact that AI won’t change you, or rather force you to change, is more unexpected to me.

Come as you are

The early promise of internet was that by giving access to knowledge to the people things were going to get better. 30 years later it is clear that the average level of culture has not really moved either direction. Most people use internet for porn (not new) and fifty shades of narcissism (not new either). Therefore it is not clear that more shared wisdom, which is the promise of AI, is going to improve the average level of wisdom in the general population.

The first thing to remember is that the wisdom provided by AI is a glorified average. And if you ask yourself who are the heros, and why they are recognized and celebrated as such, then you remark that they are never the ones just following the lines nor they are the ones that behave in an average way. Everyone looking at “Schindler’s list” imagine being Oskar (never the guard of the concentration camps). But facts (!) are rather blunt : there are not a lot of Schindlers in a population. There are far more guardians. Just by this raw remark you have to imagine that AI are going to be on the side of the guardians. Heros are the exception. Statistics, which is the formal basis of AIs, mask the exception. The whole field of statistics is built precisely to eliminate the noise. Don’t cry to loud.

Go with the flow

When I think of LLMs my first idea are those lyrics from Queens Of The Stone Age:

Outside the frame
Is what we're leaving out
You won't remember anyway

AIs come with their own frame which is their learning material (and magical tweaking done behind the curtain). But it is not clear, and even if you know it you have this tendency to not remember anyway right? The thing is: most conversations are now oriented by those AIs, in the same way that “Google search” was used to settle bar fights over facts (Juventus won 3 european cup - no it was 4). When you read an interesting thread (not just interleaving insults) chances are some (most?) arguments come from interactions with AI. The integration of Grok in X makes it very easy. Too easy. As I have told previously, a problem is that the AI will lay out things as if they were self evident. If you just read the discussion, as I just pointed to in the beginning of this essay, you will have a feeling of objectivity. Something like “numbers don’t lie” feeling, and since AI are essentially crushing numbers it must be right. Maybe they are crushing bugs ?

I have a very strange feeling about all that. My intuitive fear is that the discourse will become flat. It may sound paradoxical but it goes well with this trend “internet violence is going up”, or “memes are out of control”. It is an actual paradox by the way: The Tocqueville paradox. The less conversation are confrontational the more the slightest contradiction will become unbearable. Now if you add to the mix the fact that you didn’t believe the AI… Are you are reasonable person ?