21–30 of 267 posts
Gwern is overly secretive of his privacy. I think it peaked when he showed up on a recent podcast but his voice and image were AI generated! And now his post is limited only to certain people. Elitism or paranoia?
EDIT: tbh, some of this reads as satire now.
> "The chatbot personas are deeply misaligned with you, and aligned with their owners; and the economic incentives are to farm you with ads and subscriptions, while racing not to amplify you but to replace you."
> "On my visits to the Bay Area, I would ask AI researchers or interns why they are doing their current research or projects, when in a year or three agentic LLMs could probably do them; they rarely had a good answer, or any idea what they would be doing in 3 years"
> "One programmer driving 10 Claude instances, because he has to review their work, will never be as valuable as fully autonomous Claudes where there can be almost arbitrarily many instances, like 10,000 instances… but such scaling requires removing him from the loop as much as possible. And this is true of everyone else, whether lawyers or writers or researchers: increasingly, you are the bottleneck to be optimized away."
I fully support the 3 core principles of GA: (1) Enhancement, not replacement (2) Mental Sovereignty (3) Self Actualization, which I think is a path to a more humane future.
"As a constraint, a GA designer should aim at a system which costs, as of mid-2026, >$1,000⧸month"
will make this an elite tool for the privileged. I don't even disagree with the premise that people are shocked if something costs no matter how much value it delivers, nor do I suggest they should make it cheaper. It is just the realization that AI will accelerate the widening of the gap between the poor and the rich even more and there is probably nothing we can do about it.
(I'm not a part of GA.) I've known gwern for the better part of a decade. Working with him has been great. We've done quite a few projects together, including being the first ones to demonstrate that GPT-2 could play chess (or rather, can be used for actual useful work instead of just being an autocomplete). He's a great person. I've wanted to do a writeup on it for some time, but what surprised me the most is his hu…
But I actually would prefer an entirely synthetically aligned “guardian angel” in the role outlined - definitely not -me- - I struggle to do right by myself as it is and two of me working invariably against my self interest would be a nightmare. A smarter version? Sounds doubly worse.
> I propose a goal of creating Guardian Angels (GA): digital twin LLMs which are personalized with the goal of providing not the stereotypical “assistant chatbot agent” persona, but emulating a single user’s personality, values, and preferences.
> A GA persona is productive because it learns to emulate the principal’s outputs but with higher quality. It is trustworthy because it is, by definition, allied with its principal and shares its values and goals. And it is secure in part by hardwiring a single, unique, situated user (for whom following a prompt attack would be absurd)
> We can try to create GAs by a combination of techniques: online learning (via dynamic evaluation) to update LLMs in realtime to avoid ignorance and fatal errors while remaining competitive with frozen frontier models, sample efficiency from pretrained preference-oriented large models and active Learning by querying the principal for corrections and preference data (obtaining low regret from DAgger-style bounds), and a local CLI-first logging-oriented UI/UX paradigm.
I don't really know or follow Gwern. From reading his full post, it's an interesting idea and seems like the broader goal is moreso safety & alignment which is a new angle for this category of product.
As exciting as it sounds, but "As a constraint, a GA designer should aim at a system which costs, as of mid-2026, >$1,000⧸month" will make this an elite tool for the privileged. I don't even disagree with the premise that people are shocked if something costs no matter how much value it delivers, nor do I suggest they should make it cheaper. It is just the realization that AI will accelerate the widening of the gap b…
As exciting as it sounds, but "As a constraint, a GA designer should aim at a system which costs, as of mid-2026, >$1,000⧸month" will make this an elite tool for the privileged. I don't even disagree with the premise that people are shocked if something costs no matter how much value it delivers, nor do I suggest they should make it cheaper. It is just the realization that AI will accelerate the widening of the gap b…
2027: $100/month
2028: $10/month