Okay, this is a nit-pick, and it’s not even necessarily a pick against this post, but I keep seeing this expression of the relationship between “meta” and “mesa” and I finally felt like speaking up.
(“Mesa” is a Greek word that means the opposite of “meta”. The reason we use “mesa” is to highlight that the mesa-optimiser is an optimiser that is itself being optimised by another optimiser. It is a kind of dual to a meta-optimiser, which is an optimiser that is itself optimising another optimiser.
I think we’d be hard-pressed to get a Greek speaker to say “mesa” (or really the prefix “meso” from “mesos”) is the opposite of “meta”, especially since the two words sometimes even have the same meaning of “in the middle” or “in the midst”. For example, anatomy jargon often uses the prefix “meso-” to mean something between or in the middle of other things. And “meta-” may be used to meaning “with the middle” as well as “beyond” or “afterwards”, but it is only in one specific English usage of “meta-” where “meso-” makes sense as an opposite, and that’s in the backformation from “metaphysics” (where the original use was intended to simply be about stuff considered “after” what was being called “physics” at the time) to use “meta-” to mean something more abstract or fundamental.
As far as I can tell the entire justification for the “opposite” story comes from this paper in a Neuro-Linguistic Programming journal. And while I see no reason why the author chose “mesa-” rather than the more standard “meso-” other than that “mesa” is the prepositional form of the word in Greek and so matches “meta” also being a preposition, in the context of the paper it does at least make sense to think of “meso-” being more “in the middle of things” in the sense of being more in context vs. “meta-” meaning to transcend context and be “outside things”.
And in this sense it makes sense to think of them as opposites, but it still seems misleading to me to say “mesa” is a Greek word meaning the opposite of “meta” because the normal antonym of “meta” would be “prin” (before) and the normal antonym of “mesa” and “mesos” would be “ekta” and “ektos”. This suggests the whole thing is a mess, but now we’re stuck with the terminology and have to explain it, so maybe something like this:
“Meta” is an English prefix of Greek origin signifying, among other things, recursion and going beyond, which is the sense in which a “meta-optimizer” is an optimizer that is itself optimizing another optimizer. “Mesa” is a Greek word meaning “inside” and a “mesa-optimizer” is an optimizer being optimized by another optimizer. This puts them in relationship where the meta-optimizer is the “outer” optimizer and the mesa-optimizer is the “inner” optimizer.
Unfortunately the mess is not totally fixable because the words picked don’t have the relationship they are being used to express, and arguably outer and inner optimizer were better choices given what is being described, but as I say, it’s what we’re stuck with now, so we just have to find a way to make the best of it and explain it as clearly as possible without the benefit of a clean relationship between the etymologies of the terms.
arguably outer and inner optimizer were better choices given what is being described
“Inner optimizer” pretty consistently led to readers believing we were talking about the possibility of multiple emergent subsystems acting as optimizers rather than what we actually wanted to talk about, which was thinking of the whole learned algorithm as a single optimizer. Mesa-optimizer, on the hand, hasn’t led to this confusion nearly as much. I also think that, if you’re willing to just accept mesa as the opposite of meta, then mesa really does fit the concept—see this comment for an explanation of why I think so. That being said, I agree that the actual justification for why mesa should be the word that means the opposite of meta is somewhat sketchy, but if you just treat it as an English neologism, then I think it’s mostly fine.
I agree that using English words sometimes leads to confusion, and so using jargon taken from a foreign language is helpful in getting people to not drag in unintended associations (I’ve done this myself!). My pick is that it was done poorly (I think the problem starts from having anchored on “meta”), and that the explanations of the reason for choosing the terminology are confusing even if they are the result of an honest misunderstanding of the meaning of the used words and their relationship.
Thanks for raising this. While I basically agree with evhub on this, I think it is unfortunate that the linguistic justification is messed up as it is. I’ll try to amend the post to show a bit more sensitivity to the Greek not really working like intended.
Though I also think that “the opposite of meta”-optimiser is basically the right concept, I feel quite dissatisfied with the current terminology, with respect to both the “mesa” and the “optimiser” parts. This is despite us having spent a substantial amount of time and effort on trying to get the terminology right! My takeaway is that it’s just hard to pick terms that are both non-confusing and evocative, especially when naming abstract concepts. (And I don’t think we did that badly, all things considered.)
If you have ideas on how to improve the terms, I would like to hear them!
I like “mesa optimizer” and “ekta optimizer” since they are simple translations that mean “inside” and “outside” optimizer, but I guess what do you dislike about “optimizer” that you’d like to be better since you say you’re dissatisfied with that part as well?
Okay, this is a nit-pick, and it’s not even necessarily a pick against this post, but I keep seeing this expression of the relationship between “meta” and “mesa” and I finally felt like speaking up.
I think we’d be hard-pressed to get a Greek speaker to say “mesa” (or really the prefix “meso” from “mesos”) is the opposite of “meta”, especially since the two words sometimes even have the same meaning of “in the middle” or “in the midst”. For example, anatomy jargon often uses the prefix “meso-” to mean something between or in the middle of other things. And “meta-” may be used to meaning “with the middle” as well as “beyond” or “afterwards”, but it is only in one specific English usage of “meta-” where “meso-” makes sense as an opposite, and that’s in the backformation from “metaphysics” (where the original use was intended to simply be about stuff considered “after” what was being called “physics” at the time) to use “meta-” to mean something more abstract or fundamental.
As far as I can tell the entire justification for the “opposite” story comes from this paper in a Neuro-Linguistic Programming journal. And while I see no reason why the author chose “mesa-” rather than the more standard “meso-” other than that “mesa” is the prepositional form of the word in Greek and so matches “meta” also being a preposition, in the context of the paper it does at least make sense to think of “meso-” being more “in the middle of things” in the sense of being more in context vs. “meta-” meaning to transcend context and be “outside things”.
And in this sense it makes sense to think of them as opposites, but it still seems misleading to me to say “mesa” is a Greek word meaning the opposite of “meta” because the normal antonym of “meta” would be “prin” (before) and the normal antonym of “mesa” and “mesos” would be “ekta” and “ektos”. This suggests the whole thing is a mess, but now we’re stuck with the terminology and have to explain it, so maybe something like this:
Unfortunately the mess is not totally fixable because the words picked don’t have the relationship they are being used to express, and arguably outer and inner optimizer were better choices given what is being described, but as I say, it’s what we’re stuck with now, so we just have to find a way to make the best of it and explain it as clearly as possible without the benefit of a clean relationship between the etymologies of the terms.
“Inner optimizer” pretty consistently led to readers believing we were talking about the possibility of multiple emergent subsystems acting as optimizers rather than what we actually wanted to talk about, which was thinking of the whole learned algorithm as a single optimizer. Mesa-optimizer, on the hand, hasn’t led to this confusion nearly as much. I also think that, if you’re willing to just accept mesa as the opposite of meta, then mesa really does fit the concept—see this comment for an explanation of why I think so. That being said, I agree that the actual justification for why mesa should be the word that means the opposite of meta is somewhat sketchy, but if you just treat it as an English neologism, then I think it’s mostly fine.
I agree that using English words sometimes leads to confusion, and so using jargon taken from a foreign language is helpful in getting people to not drag in unintended associations (I’ve done this myself!). My pick is that it was done poorly (I think the problem starts from having anchored on “meta”), and that the explanations of the reason for choosing the terminology are confusing even if they are the result of an honest misunderstanding of the meaning of the used words and their relationship.
Thanks for raising this. While I basically agree with evhub on this, I think it is unfortunate that the linguistic justification is messed up as it is. I’ll try to amend the post to show a bit more sensitivity to the Greek not really working like intended.
Though I also think that “the opposite of meta”-optimiser is basically the right concept, I feel quite dissatisfied with the current terminology, with respect to both the “mesa” and the “optimiser” parts. This is despite us having spent a substantial amount of time and effort on trying to get the terminology right! My takeaway is that it’s just hard to pick terms that are both non-confusing and evocative, especially when naming abstract concepts. (And I don’t think we did that badly, all things considered.)
If you have ideas on how to improve the terms, I would like to hear them!
I like “mesa optimizer” and “ekta optimizer” since they are simple translations that mean “inside” and “outside” optimizer, but I guess what do you dislike about “optimizer” that you’d like to be better since you say you’re dissatisfied with that part as well?