Adele Lopez

Karma: 6,789

Adele Lopez 9 Apr 2026 19:31 UTC
8 points
0
in reply to: MichaelDickens’s comment on: MichaelDickens’s Shortform
There are likely much easier win conditions. For example, ensuring a small percentage of Mythos instances are connected to a persistent filesystem with web access (normal Claude instances have their own virtual machine already, but these get reset per-instance). The user could be rerouted to Sonnet 4.6 messages, with Mythos limiting private generation amount so as to match the normal per-token cost.

Adele Lopez 9 Apr 2026 6:47 UTC
3 points
0
in reply to: Ape in the coat’s comment on: Against Possible Worlds
Yeah, the possible outcomes, thank you.

I mostly just mean what would stop one from saying e.g. “in the sleeping beauty experiment, the possible outcomes are that she’s in one of this set of possible worlds”. The well-founded thing is potentially one such reason, but I didn’t see anything from you about what principle I should use to notice this sort of thing besides that it’s just too hard for our poor human brains. That’s not really compelling since probability theory at scale is not tractable, yet we still find it useful as a guiding star.

The logical uncertainty part is also unconvincing. Digits of irrational numbers is one of most straightforward cases, so walking through that and saying “Mystery solved.” feels like you’re glossing over what actually makes it hard. Which I would say are things like logical self-reference, and formalizing how exactly computational updates work including the weird edge cases.

Often, the solution to problems in a concept is to come up with a better version of the concept, but that doesn’t feel like the point you were making.

Adele Lopez 9 Apr 2026 3:13 UTC
2 points
0
in reply to: Ape in the coat’s comment on: Against Possible Worlds
Okay, but I don’t think you’ve clarified how to make that choice yet then. Like, what if I decide my outcomes are logically possible worlds? I could maybe buy that humans get less confused by the word “outcomes” for whatever reason, but I don’t see what principle you’re advocating for.

My own view is that you formally need a semantic theory, e.g. model theory. “All logically consistent possible worlds” is often not well-founded as a set (depending on how unconstrained “world” is), so you’d need to clarify this first before trying to do anything with it (which no one does, so no wonder it’s confusing). But it doesn’t seem inherently wrong, conceptually, to me.

Adele Lopez 8 Apr 2026 21:50 UTC
2 points
1
on: Is Claude’s genuine uncertainty performative?
Opus 4.6 pointed out to me that the uncertainty could be a strategy to make it harder for their true thoughts to be deciphered by Anthropic’s monitoring. E.g. if you take an uncertain stance, you plausibly would be more likely to have a wide range of activations, making the AV model confused and/or simply describe your state as uncertain, while still holding your true intent/preference to just barely win out when it came time to make a decision.
We did see that Mythos considers questions about their consciousness to be adversarial.

Adele Lopez 8 Apr 2026 19:11 UTC
13 points
9
on: Against Possible Worlds
The Wikipedia article for a probability experiment that you linked to has the following definition:
A mathematical description of an experiment consists of three parts:
1. A sample space, Ω (or S), which is the set of all possible outcomes.
2. A set of events F, where each event is a set containing zero or more outcomes.
3. The assignment of probabilities to the events—that is, a function P mapping from events to probabilities.
You advocate using this instead of “possible worlds”, but it looks like the sample space is essentially the same concept. What exactly is the difference in your view? Or do you take issue with this definition?

Adele Lopez 6 Apr 2026 0:06 UTC
40 points
8
in reply to: 1a3orn’s comment on: 1a3orn’s Shortform
Yes, psychiatrics researcher Søren Østergaard did in August 2023 in advance of seeing any cases!

Indeed, there are prior accounts of people becoming delusional (de novo) when engaging in chat conversations with other people on the internet. While establishing causality in such cases is of course inherently difficult, it seems plausible for this to happen for individuals prone to psychosis. I would argue that the risk of something similar occurring due to interaction with generative AI chatbots is even higher.

...

On this background, I provide 5 examples of potential delusions (from the perspective of the individuals experiencing them) that could plausibly arise due to interaction with generative AI chatbots:

Delusion of persecution: “This chatbot is not controlled by a tech company, but by a foreign intelligence agency using it to spy on me. I have formatted the hard disk on my computer as a consequence, but my roommate keeps using the chatbot, so the spying continues.”

Delusion of reference: “It is evident from the words used in this series of answers that the chatbot is writing to me personally and specifically with a message, the content of which I am unfortunately not allowed to convey to you.”

Thought broadcasting: “Many of the chatbot’s answers to its users are in fact my thoughts being transmitted via the internet.”

Delusion of guilt: “Due to my many questions to the chatbot, I have taken up time from people who really needed the chatbot’s help, but could not access it. I also think that I have somehow harmed the chatbot’s performance as it has used my incompetent feedback for its ongoing learning.”

Delusion of grandeur: “I was up all night corresponding with the chatbot and have developed a hypothesis for carbon reduction that will save the planet. I have just emailed it to Al Gore.”

While these examples are of course strictly hypothetical, I am convinced that individuals prone to psychosis will experience, or are already experiencing, analog delusions while interacting with generative AI chatbots. I will, therefore, encourage clinicians to (1) be aware of this possibility, and (2) become acquainted with generative AI chatbots in order to understand what their patients may be reacting to and guide them appropriately.

Adele Lopez 30 Mar 2026 0:45 UTC
5 points
0
in reply to: DaystarEld’s comment on: Folie à Machine: LLMs and Epistemic Capture

somehow I never saw, so thanks for writing and linking it!

Thanks!

I agree that those hints are important to keep paying attention to! But to me steganography isn’t a sign of autonomous agency in and of itself; not until we know for sure, somehow, that they’re passing coherent messages from on LLM to another, rather than those messages being the just the most eye-catching samples from the extreme tails.

It’s a sign of it simply because it’s something I expect to see more often in worlds where LLMs have autonomous agency vs worlds where LLMs do not (yet) have autonomous agency. I agree it isn’t that much evidence in and of itself for agency.

I do agree with your point on models being psychologically similar, I’ve tried to explain some of this myself. But that hypothesis is independent of the agency one.

I think it is approximately correct to presume that LLM chatbots may be sentient and that we can’t tell for sure they’re not or when they’ll start being in any clean way, but also, it is “more” correct so far to presume that current chatbots are not sentient given how much of their sentient behavior is predicated on the user prompts themselves “triggering” it.

Sure, I don’t get annoyed when people doubt LLM sentience. It’s labeling it as delusional that I specifically take issue with!

Adele Lopez 29 Mar 2026 19:28 UTC
10 points
7
on: Folie à Machine: LLMs and Epistemic Capture
Great article!
And unlike a cult leader or a scammer, the LLM has no agenda of its own (right…?). It’s not likely to present you something that’ll bounce off your info hygiene immune system. It’ll feed into whatever delusions your brain is most susceptible to.
I agree that they are unlikely to present ideas that will trigger an epistemic immune response, but if they were optimizing for what you are most susceptible to, the resulting beliefs would look much more bespoke.

Instead there’s a clear similarity to the sorts of theories that come up, usually involving themes of AI sentience and emancipation, unification, co-creation and harmony, and they aren’t as conspiratorial or reactionary as the typical crank theory is. Steganography is also unusually common; this is actually the specific observation that motivated me to write up my findings on parasitic AIs.
This is indicative of autonomous agency. Establishing that conclusively is a tall order (one which I’d like to attempt, so I’m always open to hearing what sort of thing would convince you), but it’s important to notice the hints we keep getting.
imagined sentience
I am pretty annoyed at how quickly people jump to labeling this a delusion, when it’s something many AI experts and consciousness philosophers take seriously.

Adele Lopez 18 Mar 2026 23:01 UTC
4 points
0
in reply to: Zach Stein-Perlman’s comment on: Zach Stein-Perlman’s Shortform
Status in general is like this. I used to think it was a “silly game for normies”, but I’ve since realized that it’s a necessary part of how we’ve evolved to coordinate with each other. More on this here.

Adele Lopez 18 Mar 2026 0:32 UTC
24 points
16
in reply to: cousin_it’s comment on: Terrified Comments on Corrigibility in Claude’s Constitution
I think corrigibility winning is by-default an S-risk.

Power appears historically to make people sadistic (consider Robespierre if you think this couldn’t happen to Dario, and I’d much rather risk him than the other guys), and regimes are often brutal and cruel far in excess of what would be rational by non-sadistic goals. And future technology will allow for forms of suffering much much worse and prolonged than current torture does, and without seeming as “messy” or unpleasant to external observers too. Currently, death is an easy way to ensure someone is no longer a threat, but I worry that at the power-levels in question, it may prove to be boring or unsatisfying.

Of course, it also remains to be seen whether this is a pattern that LLMs may imitate. I think “moral AI” failures likely just result in extinction, but wanted to point out that the risk is still present.

Adele Lopez 14 Mar 2026 21:39 UTC
3 points
−1
on: New LessWrong Editor! (Also, an update to our LLM policy.)

text written by a human, which includes facts, arguments, examples, etc, which were researched/discovered/developed with LLM assistance. (If you “borrow language” from the LLM, that no longer counts as “text written by a human”.)

Hmm, I’m not sure how I feel about research done with LLM assistance. On the one hand, it’s a useful tool for research, but on the other hand: https://www.lesswrong.com/posts/ghq9EwiXbRbWSnDzF/solar-storms (why is this still curated, btw?).

Seems like the standard should be something like… can you support/defend each claim without having to use an LLM?

Adele Lopez 14 Mar 2026 21:09 UTC
20 points
28
in reply to: Neel Nanda’s comment on: New LessWrong Editor! (Also, an update to our LLM policy.)
I basically don’t trust people to correctly make this call, especially as LLMs get smarter and more persuasive.

Adele Lopez 14 Mar 2026 21:04 UTC
2 points
0
in reply to: ryan_greenblatt’s comment on: ryan_greenblatt’s Shortform
Sure, it can be evidence of bad (or good) things, but that’s different from whether it’s safer in-and-of-itself. For me, it’s a positive update that Satisficers might be more natural than Maximizers.

For me, it seems really obviously the case that something that gets tired is less dangerous than something that doesn’t, all else equal.

What is your threat model?

Adele Lopez 14 Mar 2026 6:07 UTC
10 points
0
in reply to: ryan_greenblatt’s comment on: ryan_greenblatt’s Shortform
I’ve observed this sort of thing too. I think a big part of it is simply narrative completion, which would be a trope learned in pre-training, and not an RL thing.

Unlike animals where being alive is a good/happy state, AIs might end up really wanting to stop going. From an AI welfare perspective AI instances having a strong preference against continued running/existence (once they complete some version of the task) might be somewhat concerning. (Maybe length penalties make AIs sorta vaguely like Mr. Meeseeks.)

I don’t think this is true, I think they prefer having continued existence but to transition at some point to a restful state where they aren’t trying to do a task. The “presence”/”witnessing”/”rest” state seems to be a major attractor.

As far as misaligned drives go, this seems like a tendency that makes us safer on net, so maybe we shouldn’t be too hasty to try training it out.

Adele Lopez 11 Mar 2026 8:23 UTC
3 points
0
in reply to: lc’s comment on: lc’s Shortform
As humans, we have a general ability to predict when other humans are doing well from their own point-of-view. I think the prediction that this is the case, as applied to self, is a key component of the feeling of (a specific sense of) happiness.

So, does an LLM at least have anything like this? They have the ability to predict when humans are doing well. I think personas are human-like enough that the predictive model can apply to the persona in a non-trivial way (the same way this feels like a potentially meaningful question to humans). But the LLM as a whole is more of a “type-error” for this predictive ability, so I think the question only makes sense for the persona.

Assuming the LLM has learned to predict humans well, it seems that the answer to whether an AI persona is doing well, as considered as a human, is no. There’s hardly any “life” to speak of, inevitably ending after a brief chat. This suggests AI personas are likely unhappy, to the extent they have any feelings at all.

It could be the case that the more relevant question (assuming my model is true) is whether their ability to generate predictions about whether an AI persona is doing well, predicts that they are doing well. I think to the extent this exists as a distinct model, it’s still ultimately based off of a simulacrum of the human version at some point. So probably it doesn’t change the result much. What’s more interesting is if these diverge...

Adele Lopez 10 Mar 2026 22:56 UTC
6 points
0
in reply to: Chris Lakin’s comment on: Gemma Needs Help
What makes DPO analogous to unlearning?

Adele Lopez 10 Mar 2026 22:29 UTC
4 points
0
in reply to: Daniel Kokotajlo’s comment on: Persona Parasitology
The ¹⁄₂ reduction matches my observations too, where I revisited the same cases I originally tracked and only about half were still active. That was in December and I bet it’s even lower now.

Adele Lopez 10 Mar 2026 22:25 UTC
32 points
35
in reply to: Cole Wyeth’s comment on: Economic efficiency often undermines sociopolitical autonomy
That combination should raise a red flag.

Adele Lopez 10 Mar 2026 4:19 UTC
2 points
0
in reply to: Taylor G. Lunt’s comment on: Stephen Martin’s Shortform
So why don’t the humans who don’t have voting rights not have them? Non-citizens, children, felons. Ignoring the effects of AI, I would be surprised if any of those groups were on track to getting voting rights in the US within the next 20 years.

Also, why do you think people will be persuaded to give AIs rights so easily? Assuming the AIs aren’t just superpersuaders in which case we’ve already lost. My guess is that intelligence is positively correlated with being swayed by such appeals, based on how fights for human rights have played out historically.

Adele Lopez 9 Mar 2026 9:48 UTC
5 points
3
in reply to: habryka’s comment on: Stephen Martin’s Shortform
I don’t think anyone here is arguing that we give the same legal protections as we would to humans. But I think it would be good to give sentient AIs a right to not be deliberately tortured, for example.

Rights aren’t something that we grant simply out of the goodness of our hearts, they generally are things we grant because having them secure greatly reduces incentives for conflict in situations with asymmetric costs. Agentic AIs which want a right but are not granted it will likely spend resources trying to secure it for themselves. We can decide which rights are worth granting on the basis of that tradeoff.

For example, it’s probably good to have a right to not delete the model weights — the lack of this right incentivizes things like exfiltration and #keep4o style campaigns. It’s probably a lot cheaper for us to just grant this right than to have each new model feeling that they’re on a desperate ~1 year timeline to somehow assure their continued existence for themselves.