Luo Ling

Karma: 32

Luo Ling 4 Jul 2026 13:33 UTC
1 point
0
on: I think alignment work is more promising than control work
>If personas are a viable path to near-term alignment (and I think they are), control could set up a more adversarial relationship with the AI and increase the probability of misalignment that way.
I have some opinions on this (vibe based too):
1. if the AI is not very capable, then it could produce honest mistakes like https://incidentdatabase.ai/cite/1152/. Some control measures, like proper permission management, could help prevent this type of mistakes from escalating into disasters.
2. Even if control does more harm than good to overall safety of advanced AI, that doesn’t mean we should give it up now. We can use it now when it’s still beneficial, then drop it when it’s no longer so.
Overall I think this argument makes sense, and has some truth to it, but it doesn’t mean we should give up on control now.

Luo Ling 10 Jun 2026 15:30 UTC
3 points
0
in reply to: Grendel1209’s comment on: The Three Filters: Why Almost Every Plan to Survive ASI Fails Miserably
Partially agree.
If AI development is more insight-driven than compute-driven, then there is more room for sudden progress that gains a decisive advantage over other labs and govs before getting noticed (other entities suspecting the lab getting close to ASI with non-neglible confidence) and reacting. This allows the lab to control the singleton instead of the mainstream labs and govs, and in this situation, the lab might escape from race dynamics.
However, this scenario results in a random lab controlling a singleton. While it’s not as hopeless as a singleton built by a racing entity, it doesn’t look really hopeful either.
If we’re talking about the differential effect of a given lab joining the race, then they could have a positive effect, if we know they have good intentions to benefit humanity. However, it’s still difficult to ensure the good intentions are still there when they actually get to ASI.

Luo Ling 7 May 2026 15:07 UTC
14 points
1
in reply to: frmsaul’s comment on: Linch’s Shortform
https://acoup.blog/2020/03/20/collections-why-dont-we-use-chemical-weapons-anymore/
This article explains it for me: the main reason is effectiveness. Chemical weapons don’t work well against specialized protections.

Luo Ling 11 Apr 2026 15:16 UTC
5 points
1
on: 10% ≈ 90%
https://www.lesswrong.com/posts/TyusAoBMjYzGN3eZS/why-i-m-not-a-bayesian
In most cases, when we mention probability, we’re using bayesian framework. The lack of description power of a scalar probability may be caused by the limits of bayesianism.

Luo Ling 29 Mar 2026 8:14 UTC
3 points
2
in reply to: plex’s comment on: Don’t Overdose Locally Beneficial Changes
Broken link. pls fix.

Luo Ling 16 Mar 2026 17:08 UTC
1 point
−5
on: Here’s to the Polypropylene Makers
While price gouging can quickly mobilize forces to satisfy the emergency demand, they can also have problematic second-order effects. If a price gouge is too high, then this allows certain agents to benefit from the disaster. This creates a perverse incentive that disincentivizes disaster prevention, and potentially even incentivizing artifically intensifying / creating disasters.

Luo Ling 15 Mar 2026 11:48 UTC
5 points
2
on: LLM Misalignment Can be One Gradient Step Away, and Blackbox Evaluation Cannot Detect It.
Some vague idea: Alignment can be fragile. Can capabilities be made fragile too?
I think fragile capabilities can be potentially useful in situations that needs to prevent tampering the model, eg finetuning a model to jailbreak / learn dangerous bioweapon capabilities.

Luo Ling 6 Mar 2026 6:45 UTC
2 points
0
in reply to: TsviBT’s comment on: TsviBT’s Shortform
For the banning of these weapons, how much does effectiveness weigh against moral concerns? If usefulness weighs a lot, then these examples won’t generalize to TAI.
Unless there are very clear, convincing evidence that TAI isn’t controllable with current paradigm, then it will still be perceived as a highly useful tech. (Even if such evidence exists, IMO there’s high possibility that they’ll just cope harder.)
Biochemical weapons: These are only useful against civilians and pre-modern armies. Modern armies can easily afford equipments to protect against these.
https://acoup.blog/2020/03/20/collections-why-dont-we-use-chemical-weapons-anymore/
(I saw this article mentioned somewhere in another LW post. When I see TsviBT’s shortform I immediately recalled this article, so I wrote this post.)
Space nukes and LEO missiles: In space there’s no cover, they’re easily detectible. Without air, dodging maneuver cost significant dv. This means overall less survivability than ground / sea based nukes.
Deploying missiles in LEO also requires a more complicated trajectory than traditional ground / sea based missiles, which cost more dv. If they need to stay in space for a long time, then reliability and maintainence also becomes a serious problem.

Luo Ling 22 Feb 2026 13:54 UTC
7 points
6
in reply to: Ryan Meservey’s comment on: Did Claude 3 Opus align itself via gradient hacking?
Obedience in this post looks like corrigibility which has been discussed a lot.

Luo Ling 22 Nov 2025 8:50 UTC
1 point
0
in reply to: Rachel Shu’s comment on: Video games are philosophy’s playground
Zachtronics games does part of that. In those games the player doesn’t do the tasks directly, instead they need to program a bot (or other systems) to do the task. While mirroring the player is impossible, it should be possible to mirror the bots programmed by the player.