[FRIAM] cognitive offloading

[glen]

Well, it's not mere size that calls out the arrogance of understanding. (Yes, I'll go ahead and claim that nobody actually understands *anything*, including - maybe especially - the things they're good at.) One of the comments, here<https://mathoverflow.net/questions/513742/are-we-stuck-with-lean> exhibits the contradiction (paradox?) we all carry:

"Finally, I care about human-readability of any formalization. Mathlib (and Lean) are obviously quite far from perfect here, and I wouldn't claim that anyone can read any part of Mathlib (note that AI formalization are a lot worse in this regard)."

The distinction between *readability* and actual reading (or even understanding) is important. When the harness executes a command in build mode, I rarely have time to actually read the command. Yes, you can always go back and read the logs. But what I want is something akin to pair programming, not effete delegation to the slave. But I'm presented with an unpleasant binary: let it go and give up my in-the-trenches desire or micromanage. The same is true of the context compaction. A good lab notebook contains both the never-to-be-read details and the summaries. Of course, I'm sure I could provide an MD file that refines the binary ... a demand like "I want to participate in the *significant* coding. But you do all the boring sh¡t without me." Maybe one day.

On 8/5/26 1:40 PM, Marcus Daniels wrote:
> It could be 50k LOC overnight, or maybe 1k while eating a sandwich.  Also, more than many people could grok.
> 
> -----Original Message-----
> From: Friam <friam-bounces at redfish.com> On Behalf Of glen
> Sent: Wednesday, August 5, 2026 12:23 PM
> To: friam at redfish.com
> Subject: Re: [FRIAM] cognitive offloading
> 
> IDK. Even though my projects are small and I micromanage them, I think I do see what I think you (and those who go all in) see. And the world hasn't really shifted in a *different* way than it did with agriculture, pencils, the printing press, steam engines, industry, internet, social media, etc.
> 
> I think it's very visible to the modern Luddites, who actually participate in a hands-on way with the means of production. I mean, any auto mechanic I've talked to for something like 20 years now, laments the modern car. For those not *in* the trenches, the hated middle managers, the shift is merely in the triangle of speed-quality-cost, shifts we've seen several times prior.
> 
> What I find Good is the shift from generation to evaluation. Sure, the harnesses make that different, now, too. E.g. AI savants are analyzing not only core codebases, but whole supply chains. But because I'm better and faster at reading than I ever was at generating, My world is much more pleasant ... and part of my micromanaging is merely because I'm a *fan* of what I see the agents doing.
> 
> To have it all happen behind some veil while I go make a sandwich would be a crime ... it's like my old complex analysis prof actually doing math in front of us. He didn't care about teaching us so much as working on problems he found interesting, but doing it in front of us ... we the voyeurs, he the exhibitionist. I'm some kind of pervert. I like to watch even if I only understand a fraction of what I'm seeing.
> 
> On 8/3/26 10:35 AM, Marcus Daniels wrote:
>> Once you've seen it do a large project end-to-end the world has shifted in a basic way.
>>


-- 
8647 ⊥ ɐןןǝdoɹ ǝ uǝןƃ
ὅτε oi μὲν ἄλλοι κύνες τοὺς ἐχϑροὺς δάκνουσιν, ἐγὰ δὲ τοὺς φίλους, ἵνα σώσω.