Re: The Slopification of the CCG

Manu wrote:
"the stochastic norm" is exactly the wrong thing to do in many cases; we're
trying to bring something cohesive into existence that has not existed
before."

Just yesterday, I saw the impact of the "stochastic norm" in a "discussion"
with Claude. I asked it about the impact of the Social Security Trust
Fund's insolvency. It immediately regurgitated many pages of dire
commentary. I then asked it to answer the same question but to consider the
United States a "monetary sovereign" and to reference Abba Lerner's 1943
paper on Functional Finance. Claude then informed me that Trust Fund
solvency had absolutely no economic relevance. It assured me that
insolvency would be a purely political event and that the political
reaction would be vastly worse than the essentially non-existent economic
impact.

I then asked Claude to explain the difference. It said:

> My training data is a sample of the discourse that exists, and that
> discourse is overwhelmingly dominated by frameworks that treat the Trust
> Fund as economically meaningful. Because you asked me simply "is the Social
> Security Trust Fund useful?" without the Functional Finance framing, I
> produced a balanced-sounding answer that gave significant weight to the
> Trust Fund as a commitment device, a savings mechanism, and a guarantor of
> solvency — all of which are, from a Functional Finance standpoint, either
> category errors or political fictions dressed as economics."

... What you've identified is a specific instance of a broader failure mode
> in how I work. My *outputs tend to reflect the distribution of arguments
> in my training data, not the logical weight of those arguments*. If 90%
> of the economics commentary treats the Trust Fund as real and meaningful, I
> will tend to reproduce that treatment — *even when the 10% that
> challenges it has the stronger logical case*.


Claude then suggested that:

> Your approach in this conversation — coming in with a specific theoretical
> framework and demanding I reason within it — is actually the correct
> countermeasure. It forces me past the distributional center of my training
> data toward the logical implications of a particular set of premises.
> Without that anchoring, I drift toward the consensus, which on this topic
> is consensus built on an eighty-year-old cognitive fossil.


So, if your goal is to "bring something cohesive into existence that has
not existed before," you must be very, very careful in how you use an LLM.
Otherwise, you'll get a simple rehash of what already exists.

bob wyman


On Fri, Apr 24, 2026 at 8:49 AM Manu Sporny <msporny@digitalbazaar.com>
wrote:

> On Fri, Apr 24, 2026 at 2:01 AM Marcus Engvall <marcus@engvall.email>
> wrote:
> > The point of considered writing is to structure and formulate your ideas
> and intent well enough so that they can be effectively received,
> comprehended, and potentially acted on by your counterparty.
> ...
> > It seems to me that authors who expect their audience to use an AI to
> understand their original prose or their LLM-generated treatises have
> either abdicated responsibility of properly formulating and structuring
> their ideas for wider distribution
>
> Yes, exactly this ^^^. I've been trying to think of a way to say this
> in the current thread and Marcus has absolutely nailed it above.
>
> We spend *months to years* trying to tease out the right architecture,
> and then the words and prose to clearly articulate those concepts and
> guidance in these specifications. We argue, with respect toward one
> another, A LOT, to get there.
>
> We end up taking that much time because "the stochastic norm" is
> exactly the wrong thing to do in many cases; we're trying to bring
> something cohesive into existence that has not existed before.
>
> Well done, Marcus -- IMHO, you've identified the core of the social
> norm that is broken when LLMs are used to generate reams of content to
> make an unworkable idea look legitimate by placing window dressing
> around it.
>
> We're here for considered ideas and writing, not to hear stochastic
> parrots regurgitate old ideas.
>
> -- manu
>
> PS: I do think these stochastic parrots will evolve and overtake most,
> if not all, of us eventually... but that time is not now given what we
> seem to be collectively experiencing. Like any tool, we'll learn to
> use it better over time, norms will be established, and it might
> simultaneously provide great benefit and have the ability to destroy
> us all.
>
> --
> Manu Sporny - https://www.linkedin.com/in/manusporny/
> Founder/CEO - Digital Bazaar, Inc.
> https://www.digitalbazaar.com/
>
>