Lun

Karma: 60

Trained steering vectors may work as activation oracles

Lun22 Apr 2026 22:53 UTC

15 points

0 comments1 min readLW link

(lunnova.dev)

Lun 22 Apr 2026 20:00 UTC
9 points
0
on: Why no new notations since 1960?
Type theory?

Lun 21 Apr 2026 21:04 UTC
1 point
0
in reply to: Huluk’s comment on: Why indoor lighting is hard to get right and how to fix it
Thanks for sharing.

I’m running a GE Lighting 17259 ceramic metal halide bulb in a HANG-A-LIGHT brand fixture now, either alone or in addition to a string light with dicuno high CRI LEDs.

There’s something nice about the harsh shadows from the closer-to-point source arc than a string or large COB LED, feels a bit like I have a star in a bottle.

Sadly no superpowers or major life changes attributable to it. I think it might be slightly more effective at promoting wakefulness than my previous LED only setup but I haven’t done any proper empiricism here and have no way to distinguish whether that’s related to the spectrum or just MORE DAKKA.

Lun 21 Apr 2026 21:00 UTC
2 points
0
in reply to: David Joshua Sartor’s comment on: “You Have Not Been a Good User” (LessWrong’s second album)
For convenience, here’s what I ended up with by fetching the mp3 files and then editing the stored metadata to try to match the online version so they’d play nicely in a media library.

If rehosting isn’t ok LMK and I’ll take the shared folder down, seems unlikely it’s a problem from context.

Lun 24 Feb 2026 1:31 UTC
7 points
2
on: The persona selection model
See also: Simulators — edit: which is already cited as prior work oops

Lun 18 Nov 2025 17:46 UTC
3 points
0
in reply to: Huluk’s comment on: Why indoor lighting is hard to get right and how to fix it
Did you end up using metal halide fittings? If yes, how did that go?

After recent research finding Opsin-5 present in multiple tissues, sensitive to ≈380nm, I’m considering options that have different emission spectra from typical LED lighting to potentially shore up gaps.

Lun 11 Jul 2025 19:03 UTC
6 points
2
in reply to: Cedar’s comment on: Lun’s Shortform
fwiw I made my account in January, which I guess is still very new relative to average age of account here but hopefully means you can trust I didn’t make this account just to drop a link to the lumina post.

Lun’s Shortform

Lun11 Jul 2025 4:38 UTC

1 point

11 comments1 min readLW link

Lun 11 Jul 2025 4:38 UTC
26 points
1
on: Lun’s Shortform
Someone has posted about a personal case of vision deterioration after taking lumina and a proposed mechanism of action. I learned about lumina on lesswrong a few years back, so sharing this link.
https://substack.com/home/post/p-168042147
For the past several months I have been slowly losing my vision, and I may be able to trace it back to taking the Lumina Probiotic. Or rather, one of its byproducts that isn’t listed in the advertising
I don’t know enough about this to make an informed judgement on the accuracy of the proposed mechanism.

Lun 24 Jan 2025 10:13 UTC
1 point
0
on: Why mesa-optimization is less likely under gradient descent than natural selection
Gradient descent generates updates by suggesting algorithmic improvements for single training examples, thereby exerting much less pressure for generality than evolution does.
A recent technique Gradient Agreement Filtering filters out gradients that disagree between samples, which if I’m understanding correctly is intentionally breaking this crux and pushing for more generalization / less memorization of specific samples.

Vague intuition that with typical LM pretraining which involves batches of far more than 1 sample + optimizers with momentum this might already not hold, non-generalizing / noisy / disagreeing updates are unimportant over the training run and the generalizing, agreeing updates stick around.