10 comments

[ 1.9 ms ] story [ 33.0 ms ] thread
Lost me at “low AI exhaust from those setting technical direction should raise questions”

Wonder what Linus Torvalds’ AI exhaust is??

Linus used it to make some toy software I believe.
I've come to think that one unequivocal upside of "affordable ASI" would be to bring about a world of discourse where discursers would hesitate to think of (not to mention carry) themselves as experts.

[Perhaps it reminds me of Le Guin's thoughts on tech X power, especially as represented in The Dispossessed]

it would seem not just rude, but even snide, to bias your interlocutor towards not carefully verifying the dependencies of your exhaust

Widely dispersed and affordable AI would help because it (certainly, imho) makes the alternative, "trust but verify", so easy

And it would be strictly easier than using cloud chatGPT to sow misinformation because there are extra steps in that (--- nonsockpuppets would be inclined to verify the gist of what they are about to say as indeed misleading)

[As a bonus, this seems compatible with TFA, non-ideological anti-anti-intellectualism, and

https://engines.egr.uh.edu/episode/1495

if we default to assuming that technical leaders should be different (not more, not less) than expert engineers

Should extremist technical leaders would even strive to stay anonymous..? ]

Wow what a terrible article, just awful stuff lol
It was written by the LinkedIn Cringe bot
Do:

- LOC matter?

- Do PRs?

- Do Tokens?

I’ve never looked up a KPI to figure out who the best engineer is. It’s always obvious to everyone who the best engineer is

Natural leaders inspire. They don’t need KPIs

Great for a small startup. Now do this for 1000 people team.
> Here are some conclusions I’ve reached through experimentation:

...a bunch of stuff specific to certain models the author is currently using.

Experimentation is good for technical leaders, but you want to be focusing on approaches and understanding the fundamental constraints, not detail.

Many of the things that have completely changed in then last year will completely change in the next year.

Author here. Thanks for the comment.

I did refer to some specific models, though a lot of these learnings are from experience over the past 6 months or so, and continue to generalize as frontier models improve.

Curious if there’s a specific point you feel is too detailed?

Fair - some of the points in these examples will rot. What I think transfers are the failure modes: context rot, agents filling in decisions, whether a codebase is greppable. Those can turn into team principles, and you can only learn this through experimentation.