Posts
Comments
Yeah definitional. I think "I should do x" means about the same as "It's ethical to do x". In the latter the indexical "I" has disappeared, indicating that it's a global statement, not a local one, objective rather than subjective. But "I care about doing x" is local/subjective because it doesn't contain words like "should", "ethical", or "moral patienthood".
Ethics is a global concept, not many local ones. That I care more about myself than about people far away from me doesn't mean that this makes an ethical difference.
This seems to just repeat the repugnant conclusion paradox in more graphic detail. Any paradox is such that one can make highly compelling arguments for either side. That's why it's called a paradox. But doing this won't solve the problem. A quote from Robert Nozick:
Given two such compelling opposing arguments, it will not do to rest content with one's belief that one knows what to do. Nor will it do to just repeat one of the arguments, loudly and slowly. One must also disarm the opposing argument; explain away its force while showing it due respect.
Tailcalled talked about this two years ago. A model which predicts text does a form of imitation learning. So it is bounded by the text it imitates, and by the intelligence of humans who have written the text. Models which predict future sensory inputs (called "predictive coding" in neuroscience, or "the dark matter of intelligence" by LeCun) don't have such a limitation, as they predict reality more directly.
This still included other algorithmically determined tweets -- from what your followers had liked and later more generally "recommended" tweets. These are no longer present in the following tab.
I'm pretty sure there were no tabs at all before the acquisition.
Twitter did use an algorithmic timeline before (e.g. tweets you might be interested in, tweets people you followed liked), it was just less algorithmic than the "for you" tab currently. The time when it was completely like the current "following" tab was many years ago.
The algorithm has been horrific for a while
After Musk took over, they implemented a mode which doesn't use an algorithm on the timeline at all. It's the "following" tab.
In the past we already had examples ("logical AI", "Bayesian AI") where galaxy-brained mathematical approaches lost out against less theory-based software engineering.
Cities are very heavily Democratic, while rural areas are only moderately Republican.
I think this isn't compatible with both getting about equally many votes. Because much more US Americans live in cities than in rural areas:
In 2020, about 82.66 percent of the total population in the United States lived in cities and urban areas.
https://www.statista.com/statistics/269967/urbanization-in-the-united-states/
It's not that "they" should be more precise, but that "we" would like to have more precise information.
We know pretty conclusively now from The Information and Bloomberg that for OpenAI, Google and Anthropic, new frontier base LLMs have yielded disappointing performance gains. The question is which of your possibilities did cause this.
They do mention that the availability of high quality training data (text) is an issue, which suggests it's probably not your first bullet point.
Ah yes, the fork asymmetry. I think Pearl believes that correlations reduce to causations, so this is probably why he wouldn't particularly try to, conversely, reduce causal structure to a set of (in)dependencies. I'm not sure whether the latter reduction is ultimately possible in the universe. Are the correlations present in the universe, e.g. defined via the Albert/Loewer Mentaculus probability distribution, sufficient to recover the familiar causal structure of the universe?
This approach goes back to Hans Reichenbach's book The Direction of Time. I think the problem is that the set of independencies alone is not sufficient to determine a causal and temporal order. For example, the same independencies between three variables could be interpreted as the chains and . I think Pearl talks about this issue in the last chapter.
If base model scaling has indeed broken down, I wonder how this manifests. Does the Chinchilla scaling law no longer hold beyond a certain size? Or does it still hold, but reduction in prediction loss no longer goes along with a proportional increase in benchmark performance? The latter could mean the quality of the (largely human generated) training data is the bottle neck.
"Misinterpretation" is somewhat ambiguous. It either means not correctly interpreting the intent of an instruction (and therefore also not acting on that intent) or correctly understanding the intent of the instruction while still acting on a different interpretation. The latter is presumably what the outcome pump was assumed to do. LLMs can apparently both understand and act on instructions pretty well. The latter was not at all clear in the past.
Interesting. Question: Why does the prediction confidence start at 0.5? And how is the "actual accuracy" calculated?
I think I ger what you mean, though making more assumptions is perhaps not the best way to think about it. Logic is monotonic (classical logic at least), meaning that a valid proof remains valid even when adding more assumptions. The "taking advantage of some structure" seems to be different.
Note, the quantity you refer to is called entropy by Wikipedia, not Shannon information.
Is this a reaction to OpenAI Shifts Strategy as Rate of ‘GPT’ AI Improvements Slows?
We arguably have already colonized Antarctica. See Wikipedia.
A similar point would be: There is no permanent deep sea settlement (an underwater habitat), although this would be much easier to achieve than a settlement on Mars.
Yudkowsky has written about it:
(...) In standard metaethical terms, we have managed to rescue 'moral cognitivism' (statements about rightness have truth-values) and 'moral realism' (there is a fact of the matter out there about how right something is). We have not however managed to rescue the pretheoretic intuition underlying 'moral internalism' (...)
Replace in the post "morality" with "rationality" and you get a reductio ad absurdum.
I made basically the same proposal here, but phrased as a task of translating between a long alien message and human languages: https://www.lesswrong.com/posts/J3zA3T9RTLkKYNgjw/is-llm-translation-without-rosetta-stone-possible See also the comments, which contain a reference to a paper with a related approach on unsupervised machine translation. Also this comment echoes your post:
I think this is a really interesting question since it seems like it should neatly split the "LLMs are just next token predictors" crows from the "LLMs actually display understanding" crowd.
Arguably, "basic logical principles" are those that are true in natural language. Otherwise nothing stops us from considering absurd logical systems where "true and true" is false, or the like. Likewise, "one plus one is two" seems to be a "basic mathematical principle" in natural language. Any axiomatization which produces "one plus one is three" can be dismissed on grounds of contradicting the meanings of terms like "one" or "plus" in natural language.
The trouble with set theory is that, unlike logic or arithmetic, it often doesn't involve strong intuitions from natural language. Sets are a fairly artificial concept compared to natural language collections (empty sets, for example, can produce arbitrary nestings), especially when it comes to infinite sets.
Interesting, I really hope TMS gains more acceptance. By the way, according to studies, ECT (the precursor of TMS) is even more effective, though it does have more side effects, due to the required anesthesia, and it is gatekept even more strongly. In my youth I suffered from depression for several years, and all of this likely would have been avoidable with a few ECT sessions (TMS wasn't a thing yet), if it wasn't for the medical system's irrational bias in favor of exclusively using SSRIs and CBT. I think this happens because most medical staff have no idea how terrible depression can be, so they don't get the sense of urgency they'd get from more visible diseases.
Guys, for this specific case you really have to say what OS you are using. Otherwise you might be totally talking past each other.
(Font-size didn't change on any OS, but the font itself changed from Calibri to Gill Sans on Windows. Gill Sans has a slightly smaller x-height so probably looks a bit smaller.)
I tested it on Android, it's the same for both Firefox and Chrome. The font looks significantly smaller than the old font, likely due to the smaller x-height you mentioned. Could the font size of the comments be increased a bit so that it appears visually about as large as the old one? Currently I find it too small to read comfortably. (Subjective font size is often different from the standard font size measure. E.g. Verdana appears a lot larger than Arial at the same standard "size".)
(A general note: some people are short sighted and wear glasses, and the more short-sighted you are, the stronger the glasses contract your field of view to a smaller area. So things that may appear as an acceptable size for people who aren't particularly short-sighted, may appear too small for more short-sighted people.)
Did the font size in comments change? It does seem quite small now...
Of course for a "real" prisoners dilemma any form of coordination is ruled out from the start. But in real world instances, coordination can sometimes be introduced into systems that previously were prisoner's dilemmas. That's what I mean with "solving" a prisoner's dilemma. Making the dilemma go away.
The thing I'm pointing out here is that "coordination" is a very unspecific term, and one concrete form of coordination is being able to vote for cooperation. (Example: voting on a climate change bill instead of trying to minimize your personal carbon footprint, which would make you personally significantly worse of with hardly any benefit on the whole, which is why you would defect but vote on cooperate.) I think voting is usually not appreciated as a method of coordination, only as a method of choosing the most popular policy/party, which doesn't need to involve solving a prisoner's dilemma.
Some issues that seem to be controversial are really taboo, or arise due to an underlying taboo. For this case I have two general recommendations here.
Related to this: Some opinions may be often expressed because of virtue signalling; e.g. because the opposite is taboo, or for other reasons. Hearing such opinions doesn't provide significant testimonial evidence for their truth, since people don't hold them because of evidence they encountered, but because they feel virtuous. Though it is not easy to recognize why particular opinions are being expressed, whether they are motivated by signalling or not.
Solutions to a prisoner's dilemma are typically assumed to involve "coordination" in some sense. But what kinds of mechanism are appropriate examples for coordination? For an N-person prisoner's dilemma, one form of coordination is implementing voting. Say, everyone is forced to cooperate when the majority votes "cooperate". Nobody has a selfish interest to cooperate, but everyone has a selfish interest to vote for "cooperate".
This is interesting because economists often see voting as irrational for decision theoretic reasons. But from the game theoretic perspective above, it appears to be rational. This is probably not a new insight, but I haven't seen voting being portrayed as a type of solution to N-person prisoner's dilemmas.
One bit could also encode "probably true" and "probably false". It doesn't have to be "certainly true" and "certainly false". And this is of course what we observe. We aren't perfectly certain in everything we can barely remember to be true.
Thanks, this was an interesting article. The irony of course being that I, not knowing Russian, read it using Google Translate.
To push back a little:
In fact, I’d go further and argue that explainatory modeling is just a mistaken approach to predictive modeling. Why do we want to understand how things work? To make better decisions. But we don’t really need to understand how things work to make better decisions, we just need to know how things will react to our what we do.
The word "react" here is a causal term. To predict how things will "react" we need some sort of causal model.
What makes predictive modeling a better idea is that it also allows us to find factors that are not causal, but still useful.
Usefulness is also a causal notion. X is useful if it causes a good outcome. If X doesn't cause a good outcome, but is merely correlated with it, it isn't useful.
This might be a possible solution to the "supply-demand paradox": sometimes things (e.g. concert or soccer tickets, new playstations) are sold at a price such that the demand far outweighs the supply. Standard economic theory predicts that the price would be increased in such cases.
Ah. So I guess you should neither ask just questions, nor just always talk about yourself, but rather balance both.
Praising questions is an interesting tip to ease-in more introverted people. This reminds me of another strategy a sociable colleague of mine used on shy people: He made playful jokes about his conversation partner ("teasing"), encouraging them to be bold and "hit back" with something, and then laugh about their joke. Which instantly empowered them. This probably works best when both are men. Making self-deprecating jokes is a safer option, though not as effective.
Could you summarize a few? :)
How can you avoid that a conversation degrades into an interview? Where you are the person asking questions and the other is just answering and isn't asking anything back. (This can sometimes work, as you note in your McKinsey example, but is mostly awkward.)
I think "notice signposts" is just one of many unknown tips for good conversations, and perhaps not the most important one. Of course, people with healthy amounts of social intelligence won't need those tips anyway.
Yeah. I think the technical term for that would be cringe.
Problem is, motivated reasoning can only explain selfish beliefs, beliefs which are in accordance with our own motivations. But moral beliefs are often not at all selfish. In contrast, "suffering is bad" could just be part of what "bad" means. No motivated reasoning required. It would be a "foundational belief" in the same sense "Bachelors are unmarried" could be called "foundational".
Thoughts, @Jeremy Gillen?
If I believe eating meat is not bad because I engage in motivated reasoning, then this is, like all forms of motivated reasoning, just an irrational belief. But if I believe eating meat is not bad because I believe it doesn't create a significant amount of additional suffering, there is nothing irrational about that belief. So motivated reasoning can only explain (some) irrational beliefs. Not all beliefs about things being good or bad.
However, when something being bad means that it decreases some sort of welfare in some general way, then we don't have this problem. Now, what exactly does "welfare" etc mean? That's a question that normative ethicists try to figure out. For example via various proposed theories of utilitarianism. If philosophers are analyzing a subject matter, it's safe to assume they are analyzing some concept. Now, what's a concept? It's a meaning of a word. Like "good" or "bad".
There is a sort of opposite to assurance, where someone communicates their intention to do something (not merely a prediction that they will do it), but without creating any responsibility to follow through. This is usually done via non-verbal communication, like tone of voice, facial expression and "body language". The fact that the intention was communicated non-verbally creates plausible deniability. This happens, for example, in relationships. E.g. when asking after a date if they want to go to his/her room.
Follow-up question: If sunglasses are so cool, why do relatively few people wear them? Perhaps they aren't that cool after all?
We believe many things because we are somewhat rational; we consider hypotheses, compare them with observed evidence, and emphasise those that are more compatible. Goodness and badness defy this practice; normal reasoning does not produce hypotheses of the form "if X is morally good then I will observe Y".
I disagree. If X is an action, it is usually considered good if it increases welfare, and bad if it decreases welfare. And we can find evidence for or against something being conducive to welfare. So we can find evidence for or against something being good.
But if is "It's a cat" and is "It has four legs", and describes our beliefs (or more precisely, say, my beliefs at 5 pm UTC October 20, 2024), then . Which surely means is a materially good reason for . But , so the inference from to is still logically bad. So we don't have logicism about reasons in probability theory. Moreover, probability expressions are not invariant under substituting non-logical vocabulary. For example, if is "It has two legs", and we substitute with , then . Which can only mean the inference from to is materially bad.
Laws of probability theory still impose a structure on relations between material concepts (there are still forms of monotonicity and transitivity), whereas the logical-expressivist order of explanation argues that the theoretician isn't entitled to a priori impose such a structure on all material concepts: rather, their job is to describe them.
I think the axioms of probability can be thought of as being relative to material conceptual relations. Specifically, the additivity axiom says that the probabilities of "mutually exclusive" statements can be added together to yield the probability of their disjunction. What does "mutually exclusive" mean? Logically inconsistent? Not necessarily. It could simply mean materially inconsistent. For example, "Bob is married" and "Bob is a bachelor" are (materially, though not logically) mutually exclusive. So their probabilities can be added to form the disjunction. (This arguably also solves the problem of logical omniscience, see here).
What's the solution?
Not sure what you mean here, but would linearly transform a probability from [0..1] to [-1..1]. You could likewise transform a correlation coefficient to [0..1] with . For , this would correspond to the probability of A occuring if and only if B occurs. I.e. when .
Assigning a low probability that I will do a task in time is a self-fulfilling prophecy. Because the expected utility (probability times utility) is low, the motivation to do the task decreases. Ideally I would never assign probabilities to acts when choosing what to do, and only compare their utilities.
What's correlation space, as opposed to probability space?