Is My Toddler a Stochastic Parrot?
November 18, 2023 5:26 PM Subscribe
Is My Toddler a Stochastic Parrot? “The world is racing to develop ever more sophisticated large language models while a small language model unfurls itself in my home. It’s funny to observe the similarities between the two models.” (Single Link New Yorker, via Kottke)
That was lovely.
posted by signal at 5:46 PM on November 18, 2023 [3 favorites]
posted by signal at 5:46 PM on November 18, 2023 [3 favorites]
It's not always apparent when you're in the middle of raising them, but toddlers are basically magic.
posted by mittens at 5:57 PM on November 18, 2023 [11 favorites]
posted by mittens at 5:57 PM on November 18, 2023 [11 favorites]
The illustrations added a lot to this.
I liked the thought that it's solipsistic to think of everyone in those terms. It's true, and I didn't think it before because I think I was... focused on my own POV...
posted by Acari at 6:25 PM on November 18, 2023 [3 favorites]
I liked the thought that it's solipsistic to think of everyone in those terms. It's true, and I didn't think it before because I think I was... focused on my own POV...
posted by Acari at 6:25 PM on November 18, 2023 [3 favorites]
It’s nice to see someone else is as concerned as I am about accidentally ascribing meaning to babbles. I live my life in constant fear of mistakenly believing something good is happening, but nowhere more than in this context.
We took Tiny Monster to his 30-month checkup a few weeks ago and were asked how many words he had. His mother hesitated but I immediately said zero. He has clearly enunciated roughly a hundred words so far, almost all of them exactly once. Post, banana, dog, throw, papa, up, down, no. This morning he said goddamn. I can’t draw a line from the sound to anything he appears to be trying to express, so I can’t call them words.
I’m lost in sapirwhorfian speculation over what he can even think or remember without language with which to encode it.
On the ChatGPT front my company published its policy on using AI to generate work and I was dismayed to see that it wasn’t “don’t.” I’m not old enough to remember when CAD wiped out the drafters but if LLMs wipe out the engineers too I guess that’s just progress.
posted by The Monster at the End of this Thread at 7:14 PM on November 18, 2023 [4 favorites]
We took Tiny Monster to his 30-month checkup a few weeks ago and were asked how many words he had. His mother hesitated but I immediately said zero. He has clearly enunciated roughly a hundred words so far, almost all of them exactly once. Post, banana, dog, throw, papa, up, down, no. This morning he said goddamn. I can’t draw a line from the sound to anything he appears to be trying to express, so I can’t call them words.
I’m lost in sapirwhorfian speculation over what he can even think or remember without language with which to encode it.
On the ChatGPT front my company published its policy on using AI to generate work and I was dismayed to see that it wasn’t “don’t.” I’m not old enough to remember when CAD wiped out the drafters but if LLMs wipe out the engineers too I guess that’s just progress.
posted by The Monster at the End of this Thread at 7:14 PM on November 18, 2023 [4 favorites]
I’m not old enough to remember when CAD wiped out the drafters
I'm old enough to have seen the proof readers and copy holders lost to spellcheck
posted by mbo at 7:34 PM on November 18, 2023 [3 favorites]
I'm old enough to have seen the proof readers and copy holders lost to spellcheck
posted by mbo at 7:34 PM on November 18, 2023 [3 favorites]
concerned as I am about accidentally ascribing meaning to babbles
Our 14 month olds are at this exact “cat!” stage and it’s clear there is meaning there, but of a particularly amorphous kind, where any animal might elicit “cat”, or a random point at nothing at all may be accompanied by the same sound. I’m not at all worried about mistakenly ascribing meaning to it, though, as I’m relatively sure the feedback they get to their babbles and fumbling word usage is a large part of how they begin to construct meaning in the first place, and even if the meaning isn’t there now, responding as if it is, or what it might be, is all part of it.
posted by Jon Mitchell at 7:36 PM on November 18, 2023 [14 favorites]
Our 14 month olds are at this exact “cat!” stage and it’s clear there is meaning there, but of a particularly amorphous kind, where any animal might elicit “cat”, or a random point at nothing at all may be accompanied by the same sound. I’m not at all worried about mistakenly ascribing meaning to it, though, as I’m relatively sure the feedback they get to their babbles and fumbling word usage is a large part of how they begin to construct meaning in the first place, and even if the meaning isn’t there now, responding as if it is, or what it might be, is all part of it.
posted by Jon Mitchell at 7:36 PM on November 18, 2023 [14 favorites]
huh our 13 month old will take a list of ingredients in our fridge and confidently return a recipe for a meal that will produce approximately 3 cubic meters of chlorine gas at during its preparation.
but that's just kids i guess
posted by logicpunk at 7:58 PM on November 18, 2023 [23 favorites]
but that's just kids i guess
posted by logicpunk at 7:58 PM on November 18, 2023 [23 favorites]
I haven’t read the article yet because I’m that kind of person*, but is there a Very Big Difference between an LLM that gets trained and then used, and a toddler who is constantly experiencing the world and being socialised in the use of language? That the constant feedback loop, not only in the “this response was what I was/n’t expecting” sense, but in the “someone else understands me, and I have thoughts and feelings I want to communicate” sense - the drive to [learn to] communicate that social animals seem to have - is what makes the way people learn language and the way an LLM works not really at all comparable?
Feels like we’re comparing a very convincing marionette with the person controlling it- one of them acts like a person, but the other thinks like one. For all the layers of processing an LLM has, it’s missing the drive beneath.
Imagine a toddler who only speaks when spoken to - obviously not a toddler at all.
*A dickhead
posted by selfish at 8:35 PM on November 18, 2023 [5 favorites]
Feels like we’re comparing a very convincing marionette with the person controlling it- one of them acts like a person, but the other thinks like one. For all the layers of processing an LLM has, it’s missing the drive beneath.
Imagine a toddler who only speaks when spoken to - obviously not a toddler at all.
*A dickhead
posted by selfish at 8:35 PM on November 18, 2023 [5 favorites]
accidentally ascribing meaning to babbles
One thing I noticed is that my 22 month old (and he’s a chatty kid) understands way, way, way more than he can say. A scary amount more... You probably have noticed this too. Sometimes I’ll make a complex request to change his behavior I swear there’s no way he understands, and then he nods and changes his behavior in the complex way I requested, and I’m shocked. This kind of thing has convinced me there is a crazy amount of language stuff going on under the hood and it just gets sorted out differently for each kid and at different times, but the engine of language learning and comprehension is full on running.
I noticed that people outside the family seem to treat early talkers and late talkers differently, as if early talkers can understand a lot and late talkers can’t—but unless an SLP diagnoses a problem with receptive language for a given individual kid, kids can generally understand way more than they produce. IMO more language-meaning coding going on under the hood than you might assume…
That said, signing is really helpful! My friend learned ASL with her toddler and it helped a lot with both communication and behavior, and even language learning.
posted by oh__lol at 10:41 PM on November 18, 2023 [13 favorites]
One thing I noticed is that my 22 month old (and he’s a chatty kid) understands way, way, way more than he can say. A scary amount more... You probably have noticed this too. Sometimes I’ll make a complex request to change his behavior I swear there’s no way he understands, and then he nods and changes his behavior in the complex way I requested, and I’m shocked. This kind of thing has convinced me there is a crazy amount of language stuff going on under the hood and it just gets sorted out differently for each kid and at different times, but the engine of language learning and comprehension is full on running.
I noticed that people outside the family seem to treat early talkers and late talkers differently, as if early talkers can understand a lot and late talkers can’t—but unless an SLP diagnoses a problem with receptive language for a given individual kid, kids can generally understand way more than they produce. IMO more language-meaning coding going on under the hood than you might assume…
That said, signing is really helpful! My friend learned ASL with her toddler and it helped a lot with both communication and behavior, and even language learning.
posted by oh__lol at 10:41 PM on November 18, 2023 [13 favorites]
is there a Very Big Difference between an LLM that gets trained and then used, and a toddler who is constantly experiencing the world and being socialised in the use of language?
Yep. A few occur to me.
First is that although using a LLM is computationally expensive, training one is insanely computationally - and therefore energetically, and therefore economically - expensive. This translates to almost all of what an LLM is used for having no effect on its model; unlike toddlers, LLMs do not learn and improve simply from being exercised. Releasing an improved language model is roughly analogous to building a whole new toddler from scratch.
Second is that LLMs get trained on pre-filtered, pre-selected instances of language, not on exposure to language as but part of the entire world of disparate experience that toddlers make sense of. Any feature of any LLM that even vaguely resembles an "understanding" of the real world got carried in there as a side effect of language's ability to represent it, and can therefore only ever be a pale and spotty shadow of the world models that underlie and motivate language's primary users and inventors i.e. us. A toddler can touch a hot stove but a LLM cannot.
But the really big one is that a toddler is an autonomous, self-directed entity with a direct stake in acquiring a world model good enough to let it survive to reproductive age, while a language model is an engineered artifact whose ongoing existence depends not on physical competence but rather the whims of its operators.
It's really important, when thinking about how "AI" is or is not likely to shape our future, to remember that none of these tools has any kind of existence independent of the people and corporations who run the data centres that host them, and that understanding the behaviour and motivations of those people and corporations cannot be allowed to take a back seat to telling ourselves pretty, distracting Just So stories about arguable personhood or presumed motivations of "AI" in and of themselves.
LLMs are a technology. Like all technology, they massively amplify personal power but do nothing whatsoever to amplify the likelihood that their wielders will do so in conscientious, responsible, well-informed, well-considered ways. That is their main hazard.
posted by flabdablet at 10:53 PM on November 18, 2023 [37 favorites]
Yep. A few occur to me.
First is that although using a LLM is computationally expensive, training one is insanely computationally - and therefore energetically, and therefore economically - expensive. This translates to almost all of what an LLM is used for having no effect on its model; unlike toddlers, LLMs do not learn and improve simply from being exercised. Releasing an improved language model is roughly analogous to building a whole new toddler from scratch.
Second is that LLMs get trained on pre-filtered, pre-selected instances of language, not on exposure to language as but part of the entire world of disparate experience that toddlers make sense of. Any feature of any LLM that even vaguely resembles an "understanding" of the real world got carried in there as a side effect of language's ability to represent it, and can therefore only ever be a pale and spotty shadow of the world models that underlie and motivate language's primary users and inventors i.e. us. A toddler can touch a hot stove but a LLM cannot.
But the really big one is that a toddler is an autonomous, self-directed entity with a direct stake in acquiring a world model good enough to let it survive to reproductive age, while a language model is an engineered artifact whose ongoing existence depends not on physical competence but rather the whims of its operators.
It's really important, when thinking about how "AI" is or is not likely to shape our future, to remember that none of these tools has any kind of existence independent of the people and corporations who run the data centres that host them, and that understanding the behaviour and motivations of those people and corporations cannot be allowed to take a back seat to telling ourselves pretty, distracting Just So stories about arguable personhood or presumed motivations of "AI" in and of themselves.
LLMs are a technology. Like all technology, they massively amplify personal power but do nothing whatsoever to amplify the likelihood that their wielders will do so in conscientious, responsible, well-informed, well-considered ways. That is their main hazard.
posted by flabdablet at 10:53 PM on November 18, 2023 [37 favorites]
Second is that LLMs get trained on pre-filtered, pre-selected instances of language, not on exposure to language as but part of the entire world of disparate experience that toddlers make sense of.
There are, however, multimodal models - for example the latest releases of GPT-4 can accept an image as input. Still rather narrower than the experience of being embodied, obviously, and still pre-trained on massive amounts of data rather than continuously learning.
posted by atoxyl at 11:11 PM on November 18, 2023 [2 favorites]
There are, however, multimodal models - for example the latest releases of GPT-4 can accept an image as input. Still rather narrower than the experience of being embodied, obviously, and still pre-trained on massive amounts of data rather than continuously learning.
posted by atoxyl at 11:11 PM on November 18, 2023 [2 favorites]
Sometimes I’ll make a complex request to change his behavior I swear there’s no way he understands, and then he nods and changes his behavior in the complex way I requested
Have you considered this may simply be confirmation bias on your part? My guess is there are many, many more occasions where the kid's behaviour fails to match your instructions, but that you're ignoring these in favour of the odd random incident when he (quite coincidentally) seems to be obeying you.
posted by Paul Slade at 12:36 AM on November 19, 2023 [2 favorites]
Have you considered this may simply be confirmation bias on your part? My guess is there are many, many more occasions where the kid's behaviour fails to match your instructions, but that you're ignoring these in favour of the odd random incident when he (quite coincidentally) seems to be obeying you.
posted by Paul Slade at 12:36 AM on November 19, 2023 [2 favorites]
We took Tiny Monster to his 30-month checkup a few weeks ago and were asked how many words he had. His mother hesitated but I immediately said zero. He has clearly enunciated roughly a hundred words so far, almost all of them exactly once. Post, banana, dog, throw, papa, up, down, no. This morning he said goddamn. I can’t draw a line from the sound to anything he appears to be trying to express, so I can’t call them words.
I don't think that matters for the purposes of the clinician because they are measuring the behaviour against typical peers, rather than adult comprehension. The enunciation, separate from the expression, may be of developmental significance.
posted by Ballad of Peckham Rye at 12:53 AM on November 19, 2023 [6 favorites]
I don't think that matters for the purposes of the clinician because they are measuring the behaviour against typical peers, rather than adult comprehension. The enunciation, separate from the expression, may be of developmental significance.
posted by Ballad of Peckham Rye at 12:53 AM on November 19, 2023 [6 favorites]
Children, even toddlers, understand a lot more than adults usually give them credit for. I was already of that belief before I had a niece I saw on a regular basis, but she has been strongly confirming that belief for a couple of years now.
LLMs can be surprising in their use of language at times, but they clearly have no understanding, hence the complete hallucinations they often produce. One of the things I like about Bard is that it will show you alternate responses that it generated before selecting the highest scoring alternative. The variations make it quite clear that it often has difficulty "understanding" even relatively simple prompts, much less anything about its training corpus.
That shouldn't be surprising since the LLMs weren't built to have any more understanding than is necessary to somewhat passably translate simple sentences between various languages. What looks like understanding is only the statistical analysis necessary to differentiate uses of words like "read" using nearby context. I consider it quite an amazing accomplishment, actually, but it is often cast as being much more than it really is. It's an important piece of the puzzle in making something that can be credibly called an AI, but just feeding ever larger slightly tweaked translator LLMs ever more data isn't going to get the job done.
posted by wierdo at 2:29 AM on November 19, 2023 [6 favorites]
LLMs can be surprising in their use of language at times, but they clearly have no understanding, hence the complete hallucinations they often produce. One of the things I like about Bard is that it will show you alternate responses that it generated before selecting the highest scoring alternative. The variations make it quite clear that it often has difficulty "understanding" even relatively simple prompts, much less anything about its training corpus.
That shouldn't be surprising since the LLMs weren't built to have any more understanding than is necessary to somewhat passably translate simple sentences between various languages. What looks like understanding is only the statistical analysis necessary to differentiate uses of words like "read" using nearby context. I consider it quite an amazing accomplishment, actually, but it is often cast as being much more than it really is. It's an important piece of the puzzle in making something that can be credibly called an AI, but just feeding ever larger slightly tweaked translator LLMs ever more data isn't going to get the job done.
posted by wierdo at 2:29 AM on November 19, 2023 [6 favorites]
A thing I've noticed with my grandkids is that they reach a stage around 2 or 3 when they don't just repeat words. They start to invent ideas and come up with combinations of words that are surprising and unique. It's when you realise that this isn't just a pooping eating machine. There is an intelligence in there that's capable of creating something entirely original. I've yet to be convinced that a machine that produces word strings in a way that makes sense to us is intelligent.
A truck full of hay bales passed our house one day. My 2 and a half year old grand daughter pointed at it and said "Moo noms".
posted by night_train at 4:04 AM on November 19, 2023 [8 favorites]
A truck full of hay bales passed our house one day. My 2 and a half year old grand daughter pointed at it and said "Moo noms".
posted by night_train at 4:04 AM on November 19, 2023 [8 favorites]
See, that's just her sheltered upbringing. If she'd seen them lying about in the paddocks before they got collected and put on the truck, she'd know they were UFO poos.
Never mind the crop circles, it's crop cylinders we should be concerned about.
posted by flabdablet at 6:17 AM on November 19, 2023 [2 favorites]
Never mind the crop circles, it's crop cylinders we should be concerned about.
posted by flabdablet at 6:17 AM on November 19, 2023 [2 favorites]
is there a Very Big Difference between an LLM that gets trained and then used, and a toddler who is constantly experiencing the world and being socialised in the use of language
There are many, many, differences, but perhaps the most blunt one is that a current generation LLM encounters *at least* something like 4 orders of magnitude (GPT-3 got ~200 billion, Chinchilla 1.4 trillion) more words of language during training than a human child does while learning a language (less than 100 million). Moreover, no one knows in the slightest how or even if current generation LLM tech could be achieved with child-scale data.
(This is the idea behind the BabyLM challenge, which is also where I pulled those stats from; it doesn't look like the results of that challenge are out yet though; the conference it is being presented at is in a few weeks so perhaps we'll find out if someone does now know. Though the format of the data used in this challenge is still extremely non-naturalistic, even if the size is aiming at naturalism.)
posted by advil at 6:55 AM on November 19, 2023 [7 favorites]
There are many, many, differences, but perhaps the most blunt one is that a current generation LLM encounters *at least* something like 4 orders of magnitude (GPT-3 got ~200 billion, Chinchilla 1.4 trillion) more words of language during training than a human child does while learning a language (less than 100 million). Moreover, no one knows in the slightest how or even if current generation LLM tech could be achieved with child-scale data.
(This is the idea behind the BabyLM challenge, which is also where I pulled those stats from; it doesn't look like the results of that challenge are out yet though; the conference it is being presented at is in a few weeks so perhaps we'll find out if someone does now know. Though the format of the data used in this challenge is still extremely non-naturalistic, even if the size is aiming at naturalism.)
posted by advil at 6:55 AM on November 19, 2023 [7 favorites]
odd random incident when he (quite coincidentally) seems to be obeying you
Well, I would say this actually happens about 50-80% of the time, and usually when he’s not emotional (which is when communication breaks down). We also have a talker who’s ahead of the curve so my best guess is that his receptive language is also comparatively ahead, but from hearing from my still-active baby bumper group I would say most people’s kids are way, way ahead in terms of receptive language vs what they produce (and one criteria SLPs use to get kids signed up for early intervention is whether their receptive language is also behind, iirc).
Actually, one way you can kinda see from experience that language isn’t entirely necessary to encode concepts a la Sapir-Whorf is that toddlers will remember and refer to things they experienced when preverbal. One of my friends told me about how her toddler asked for the penguins when on a flight. The last time they were on a flight was months earlier, when they still only had a few words (penguin not included)… when they watched a penguin movie. Other obvious thought is that toddlers who learn to sign due to a speech delay have much more sign vocabulary than spoken vocabulary, so speech is definitely not the only signifier of language acquisition and comprehension.
Magda Gerber, who was a pioneer of early childcare strategies, advocated for talking to even your newborn like they can truly understand you, and so we talked to our kid like he could understand everything we said from day one. And I get it, because they are eerily able to comprehend more than you’d think from early on. I’m sure a lot of it is them picking up on tone… they understand that we are taking the effort to explain something to them so they are more cooperative, etc. But still, it’s interesting stuff. I’m loving this time in their development and what it reveals about how people work.
posted by oh__lol at 7:36 AM on November 19, 2023 [3 favorites]
Well, I would say this actually happens about 50-80% of the time, and usually when he’s not emotional (which is when communication breaks down). We also have a talker who’s ahead of the curve so my best guess is that his receptive language is also comparatively ahead, but from hearing from my still-active baby bumper group I would say most people’s kids are way, way ahead in terms of receptive language vs what they produce (and one criteria SLPs use to get kids signed up for early intervention is whether their receptive language is also behind, iirc).
Actually, one way you can kinda see from experience that language isn’t entirely necessary to encode concepts a la Sapir-Whorf is that toddlers will remember and refer to things they experienced when preverbal. One of my friends told me about how her toddler asked for the penguins when on a flight. The last time they were on a flight was months earlier, when they still only had a few words (penguin not included)… when they watched a penguin movie. Other obvious thought is that toddlers who learn to sign due to a speech delay have much more sign vocabulary than spoken vocabulary, so speech is definitely not the only signifier of language acquisition and comprehension.
Magda Gerber, who was a pioneer of early childcare strategies, advocated for talking to even your newborn like they can truly understand you, and so we talked to our kid like he could understand everything we said from day one. And I get it, because they are eerily able to comprehend more than you’d think from early on. I’m sure a lot of it is them picking up on tone… they understand that we are taking the effort to explain something to them so they are more cooperative, etc. But still, it’s interesting stuff. I’m loving this time in their development and what it reveals about how people work.
posted by oh__lol at 7:36 AM on November 19, 2023 [3 favorites]
Never mind the crop circles, it's crop cylinders we should be concerned about.
They might have been square bales, I can't remember. Maybe robot poo?
posted by night_train at 7:38 AM on November 19, 2023
SPACE WOMBATS
posted by flabdablet at 7:45 AM on November 19, 2023 [5 favorites]
posted by flabdablet at 7:45 AM on November 19, 2023 [5 favorites]
Now who's the stochastic parrot, hmm?
posted by flabdablet at 7:46 AM on November 19, 2023 [3 favorites]
posted by flabdablet at 7:46 AM on November 19, 2023 [3 favorites]
I just want to point out that ELIZA was a laughably primitive chat bot from the 1960s, but this didn't stop it from fooling people.
I think about this a lot.
posted by AlSweigart at 11:32 AM on November 19, 2023 [7 favorites]
I think about this a lot.
posted by AlSweigart at 11:32 AM on November 19, 2023 [7 favorites]
One thing I noticed is that my 22 month old (and he’s a chatty kid) understands way, way, way more than he can say. A scary amount more...
Every so often, there's a post on Reddit's language learning subreddit to the effect of "What's wrong with me, I can understand the language I'm learning much better than I can speak it," or "Is this normal? I can understand the language I'm learning much better than I can speak it." And - I suspect it's pretty universal, both in first language acquisition and second+ language acquisition, that understanding vastly precedes production.
(I am, at the moment, reading one of the major Japanese novelists of the early 20th century; I am also certain that the next time I go to Japan, I'll have at least half a dozen embarrassing language slip-ups.)
posted by Jeanne at 2:35 PM on November 19, 2023 [2 favorites]
Every so often, there's a post on Reddit's language learning subreddit to the effect of "What's wrong with me, I can understand the language I'm learning much better than I can speak it," or "Is this normal? I can understand the language I'm learning much better than I can speak it." And - I suspect it's pretty universal, both in first language acquisition and second+ language acquisition, that understanding vastly precedes production.
(I am, at the moment, reading one of the major Japanese novelists of the early 20th century; I am also certain that the next time I go to Japan, I'll have at least half a dozen embarrassing language slip-ups.)
posted by Jeanne at 2:35 PM on November 19, 2023 [2 favorites]
Former dev psych PhD student here. In terms of language development, comprehension almost always outpaces production. I'm not familiar with the second language learning literature, but from my own n=1 experiment of relearning Spanish, I understand a lot more than what I can say.
posted by kathrynm at 11:22 AM on November 21, 2023
posted by kathrynm at 11:22 AM on November 21, 2023
We're a bilingual household - German and English. I've always spoken English with my daughter. My wife has always spoken German with her.
We live in Austria, so it's not surprising that my daughter started speaking German, first, but I always felt that her understanding of English was roughly on par with her understanding of German.
We flew to Texas to visit family and friends over the Christmas holidays shortly before my daughter turned 3. In the preceding months, I had encouraged her repeatedly to practice speaking English with me, to no avail. I explained to her that no one in Texas would understand her if she only spoke German, but my entreaties were ignored. She had no interest in speaking English with me.
My mom picked us up at the airport in Austin, and we stopped for dinner at Chuy's.
That's when my daughter started speaking English, first with my mom during dinner. She spoke English for three straight weeks, and she has spoken English with me ever since.
posted by syzygy at 2:29 PM on November 23, 2023 [1 favorite]
We live in Austria, so it's not surprising that my daughter started speaking German, first, but I always felt that her understanding of English was roughly on par with her understanding of German.
We flew to Texas to visit family and friends over the Christmas holidays shortly before my daughter turned 3. In the preceding months, I had encouraged her repeatedly to practice speaking English with me, to no avail. I explained to her that no one in Texas would understand her if she only spoke German, but my entreaties were ignored. She had no interest in speaking English with me.
My mom picked us up at the airport in Austin, and we stopped for dinner at Chuy's.
That's when my daughter started speaking English, first with my mom during dinner. She spoke English for three straight weeks, and she has spoken English with me ever since.
posted by syzygy at 2:29 PM on November 23, 2023 [1 favorite]
My mom picked us up at the airport in Austin, and we stopped for dinner at Chuy's.
That's when my daughter started speaking English, first with my mom during dinner.
Obligatory Henning Wehn bit
posted by flabdablet at 9:03 PM on November 23, 2023 [1 favorite]
That's when my daughter started speaking English, first with my mom during dinner.
Obligatory Henning Wehn bit
posted by flabdablet at 9:03 PM on November 23, 2023 [1 favorite]
« Older Badoraito o itadakimasu. | Saving this tree species from extinction Newer »
This thread has been archived and is closed to new comments
posted by clavdivs at 5:38 PM on November 18, 2023 [1 favorite]