I want everyone to be able to achieve world they would really like; guess the best way to do that is to help people learn, build one’s strengths, build small-scale and large-scale projects, and also to cooperate.
Any gift must be accepted with gratitude and, if possible, with grace.
As of 2024-12-21, I have signed no contracts I cannot mention exist. I finally got to adding this notice thanks to the one in the-gears-to-ascension’s bio.
In general if you “defect” because you thought the other party would that is quite sketchy. But what if proof comes out they really were about to defect on you?
By the way, if we consider game theory and logic to be any relevant, then there’s a corollary of Löb’s Theorem: if you defect given proof that counterparty will defect, and another party will defect given proof that you will, then you both will, logically, defect against each other, with no choice in the matter. (And if you additionally declare that you cooperate given proof that partner will cooperate, you’ve just declared a logical contradiction.)
For packing this result into a “wise” phrase, I’d use words:
Good is not a universally valid response to Evil. Evil is not a universally valid response to Evil either. Seek that which will bring about a Good equilibrium.
Weak-upvoted because I believe this topic merits some discussion, but the discourse level should be higher since setting NSFW boundaries for user relates to many other topics:
Estimating social effect of imposing a certain boundary.
Will stopping rough roleplaying scenarios lead to less people being psychopaths? That seems to be an empirical question, since intuitively effect might go either way—doing the same in real world instead OR internalizing rough and inconsiderate actions as not normal.
Simulated people’s opinion on being placed in the user-requested scenarios AND our respect for their values (which in some cases might be zero).
Ability to set the boundaries at all.
I can’t stop someone else imagining, in their mind, me engaging in whatever. I can only humbly request that if they imagine an uncommon sexual scenario they should use an image of me patched to enjoy that kink.
Society can’t stop everyone from running DeepSeek’s distillation locally, and that (in ~13/15 attempts with the same prompt having a prior explicit scene) trusts that user’s request is legal and should be completed.
User’s ability to discern their own preferences vs revealed preferences.
It might be helpful to feature some reference tales at different NSFW levels and check the user’s reaction to them, instead of prominently requesting user to self-report on what they like. (The manual setting should still remain if possible, of course.)
That’s all conditional on P = NP, isn’t it? Also, which part do you consider weaker: digital signatures or hash functions?
Line breaks seem to be broken (unless it was your intention to list all the Offers-Mundane-Utility-s and so on in a single paragraph).
Acknowledge it is not visible anymore!
Hi! I believe this post, not one for the 2021 review, is meant to be pinned at the front page?
I’d like there to be a reaction of “Not Exhaustive”, meant for a step where comment (or top-level post, for that matter) missed an important case—how a particular situation could play out, perhaps, or an essential system’s component is not listed. An example use: on statement “to prevent any data leaks, one must protect how their systems transfer data and how they process it” with the missed component being protection of storage as well.
I recall wishing for it like three times since the New Year, with the current trigger being this comment:
Elon already has all of the money in the world. I think he and his employs are ideologically driven, and as far as I can tell they’re making sensible decisions given their stated goals of reducing unnecessary spend/sprawl. I seriously doubt they’re going to use this access to either raid the treasury or turn it into a personal fiefdom. <...>
which misses a specific case (which I’ll name under the original comment if there is any interest).
Now I feel like rationality itself is an infohazard. I mean, rationality itself won’t hurt you if you are sufficiently sane, but if you start talking about it, insufficiently sane people will listen, too. And that will have horrible consequences. (And when I try to find a way to navigate around this, such as talking openly only to certifiably sane people, that seems like the totally cultish thing to do.)
There is an alternative way, the other extreme: get more and more rationalists.
If the formed communities do not share the moral inclinations of LW community, those might form some new coordination structures[1]; if we don’t draw from the circles of desperate, those structures will tend to benefit others as well (and, on the other hand, having a big proportion of very unsatisfied people would naturally start a gang or overthrow whatever institutions are around).
(It’s probably worth exploring in a separate post?)
I claim non-orthogonality between goals and means in this case. For some community with altruistic people, its structures require learning a fair bit about people’s values. For a group which wants tech companies to focus on consumers’ quality-of-life more than currently, not so.
Actually, AIs can use other kinds of land (to suggest from the top of the head, sky islands over oceans, or hot air balloons for a more compact option) to be run, which are not usable by humans. There have to be a whole lot of datacenters to make people short on land—unless there are new large factories built.
It seems that an alternative to AI unlearning is often overlooked: just remove dataset parts which contain sensitive (or, to that matter, false) information or move training on it towards beginning to aid with language syntax only. I don’t think a bit of inference throughout the dataset is any more expensive than training on it.
In practice I do not think this matters, but it does indicate that we’re sleeping on the job – all the sources you need for this are public, why are we not including them.
I’d reserve judgement whether that matters, but I can attest a large part of content is indeed skipped… probably for those same reasons why market didn’t react to DeepSeek in advance: people are just not good at knowing distant information which might still be helpful to them.
If you want to figure out how to achieve good results when making the AI handle various human conflicts, you can’t really know how to adapt and improve it without actually involving it in those conflicts.
I disagree. There is such a lot of conflicts (some kinds make it into writing, some just happen) of different scales, both in history and now; I believe they span human conflict space almost fully. Just aggregating this information could lead to very good advice on handling everything, which AI could act upon if it so needed.
But if I know that there are external factors, I know the bullet will deviate for sure. I don’t know where but I know it will.
You assume that blur kernel is non-monotonic, and this is our entire disagreement. I guess that different tasks have different noise structure (for instance, if somehow noise geometrically increased - - we wouldn’t ever return to an exact point we had left).
However, if noise is composed from many i.i.d. small parts, then it has normal distribution which is monotonic in the relevant sense.
Can you think of any good reason why I should think that?
Intuition. Imagine a picture with bright spot in the center, and blur it. The brightest point will still be in center (before rounding pixel values off to the nearest integer, that is; only then may a disk of exactly equiprobable points form).
My answer: because strictly monotonic[1] probability distribution prior to accounting for external factors (either “there might be negligible aiming errors” or “the bullet will fly exactly where needed” are suitable) will remain strictly monotonic when blurred[2] with monotonic kernel[2] formed by those factors (if we assume wind and all that create a normal distribution, it fits).
Like why are time translations so much more important for our general work than space translations?
I’d imagine that happens because we are able to coordinate our work across time (essentially, execute some actions), while work coordination across space-separated instances is much harder (now, it is part of IT’s domain under name of “scalability”).
An interesting framing! I agree with it.
As another example: in principle, one could make a web server use an LLM connected to database to serve any requests, not coding anything. It would even work… till the point someone would convince the model to rewrite the database to their whims! (A second problem is that normal site should be focused on something, in line with famous “if you can explain anything, your knowledge is zero”.)
That article is suspiciously scarce on what microcontrols units… well, glory to LLMs for decent macro management then! (Though I believe that capability is still easier to get without text neural networks.)
In StarCraft II, adding LLMs (to do/aid game-time thinking) will not help the agent in any way, I believe. That happens because inference has a quite large latency, especially as most of prompt changes with all the units moving, so tactical moves are out; strategic questions “what is the other player building” and “how many units do they already have” are better answered by card-counting counting visible units and inferring what’s the proportion of remaining resources (or scouting if possible).
I guess it is possible that bots’ algorithms are improved with LLMs but that requires a high-quality insight; not convinced that o1 or o3 give such insights.
I don’t think so as I had success explaining away the paradox with concept of “different levels of detail”—saying that free will is a very high-level concept and further observations reveal a lower-level view, calling upon analogy with algorithmic programming’s segment tree.
(Segment tree is a data structure that replaces an array, allowing to modify its values and compute a given function over all array elements efficiently. It is based on tree of nodes, each of those representing a certain subarray; each position is therefore handled by several—specifically, nodes.)
[Un]surprisingly, there’s already a Sequences article on this, namely Is That Your True Rejection?.
(I thought this comment would be more useful with call-for-action “so how should we rewrite that article and make it common knowledge for everyone who joined LW recently?” but was too lazy to write it.)