While I sort through whatever is happening with GPT-4, today’s scheduled post is two recent short stories about restaurant selection. Ye Olde Restaurante Tyler Cowen says that restaurants saying ‘since year 19xx’ are on net a bad sign, because they are frozen in time, focusing on being reliable.
He says a little about his own approach to morality in the New and Old EA Cause Areas section at https://thezvi.substack.com/p/book-review-going-infinite. It seems like he doesn't think of value as a fully systematizible thing, and instead aims to use consequentialism as a heuristic (possibly not the right word) for identifying tradeoffs that seem like they are worth taking. As opposed to thinking of consequentialism as a guide to tradeoffs that always ought to be made, given a certain level of confidence about a certain number of somethings.
Let's apply the utility monster thought experiment to a hypothetical superintelligence. Wouldn't a superintelligence that was capable of killing all humans probably also be capable of making so many copies of itself that the value-maximizing action would be killing off humans to a) use their atoms and b) prevent them from turning the superintelligence(s) off? In this scenario, do you find that you, too - for ultimately 'irrational' reasons - don't want to treat value as a fully systematizible thing?
If so, you might find it a little easier to see why Zvi would also opt out of trading off human interests against animal suffering.
I don't mean this to be combative, but isn't the most likely reason nothing but the main: that animals are delicious, and Zvi has concluded that the benefits he receives from eating animals is more important to him than the suffering the animals undergo as a result?
First of all, I don't know how Yelp is still useable. They've long since been hiding reviews (as "unreliable") an directing you to use the app, which is annoying enough that I barely use them, except if I'm really digging for information on a place.
To me, internet "ruining" things is making them popular, so the demand makes it too expensive, or ruins the quality, or the vibe. A similar thing is the case with "undiscovered" places and anything that's a good deal. These exist but not on the easily accessible public Internet.
I agree Yelp reviews do not seem helpful. I do have quite a bit of success with Google Maps ratings, but I’ve found it takes some calibration for the specific area. Some areas the baseline review is a 3.5 and a 4.5 has a good chance of being great, while other areas the baseline seems to be 4 or 4.5 and you have to be more discriminating. At the high end, the signal starts to get lost when the baseline rating is around 4.5 and I have to rely more on other methods.
So the best piece of restaurant advice I got from Tyler Cowen, in An Economist Eats Lunch, was to look for small, family owned restaurants likely subsidized by a larger business, because they're likely to be both cheap and extremely good.
His *specific* advice was to actually look for a Thai or Vietnamese place literally attached to a small motel, because that's a clear signal of that. Mine is the generalized advice I abstracted from his specific suggestion.
It served me well on a recent ~4k mile road trip across the USA, and I found numerous "grandma's recipe" style places that were indeed both cheap and excellent.
Other examples besides motel / lodging would be gas station, jerky store, butcher, winery, and farm. And I'll happily double down on "farm," if you can find a restaurant run by a farming family, go even if it's an hour round trip out of the way and the hours are terrible, because the recipes, execution, and ingredients will all be top notch in my experience.
An extreme version of this idea - When I was in high school Toyota built a factory in my hometown. A nice sushi restaurant opened up right next to it and would sell expensive sake to the Toyota execs and very reasonably priced lunch specials to anyone else. Still some of the best sushi I've had outside of Japan.
I did find that Yelp reviews were far more useful when I lived on the west coast, but that was 5-10 years ago. I don't know whether the population of users leans more hipster there, or if the service has just declined in general over that time. I would believe either or both.
There's something to the 'hipsters not tourists' point, but I don't *exactly* agree with it. Rather, one should indeed look at who is doing the rating.
I like South Asian food quite a lot, but the ratings on an Indian place in a mostly white neighbourhood/town are completely useless, whereas the rating for a place in an extremely South Asian neighbourhood (e.g. Green Street in London, Brick Lane too touristy now) are in my experience highly reliable.
Ideally the reviews on a Chinese place (assuming you're in the mood for Chinese-Chinese, not American-Chinese, noting both have their place) aren't even in English and you need to translate them.
Pub reviews are tough because ideally I want them split on food/drink/vibe, as the three are largely uncorrelated with each other and it really depends what I'm in the mood for.
Yep, there's definitely some of this. Asian restaurant ratings are by far the least reliable, many of my favorite places there don't review well at all.
Yelp is unusable as far as I can tell. Other ratings seem to select for something just beyond mediocrity. You'll see chains that are good at consistently producing a predictable, average or slightly above average rated just slightly behind what a better quality non-chain with similar pricing. A non-chain that's higher quality and higher pricing will be even closer to the chain because it gets judged more harshly by the people that are looking for good food and gets dinged for prices by people that don't appreciate quality improvements beyond that produced by chains.
Ratings in heavily tourist areas do seem to be almost worthless. A common theme for the vacation areas we live near is for a place to open up and be good, then get popular, and then either get overwhelmed by the volume and stop doing what they do well, or more typically, they take more and more shortcuts and use more and more pre-prepared foods to handle the extra volume. The second approach barely hurts the restaurants reviews for years. You will see tourists leaving those traps proudly proclaiming that the food they just ate that was basically moved off a sysco truck to a freezer, and then to an oven, griddle, grill, or fryer with no additional work necessary other than (possibly) thawing, as the best they've ever had of that type of food. Which is fine. I'm glad the combination of their expectations and good vibes from vacation produce an enjoyable experience for them. It does sort of suck that new places that are good get discovered and overwhelmed much more quickly, but I doubt the owners of those restaurants share my opinion on that though.
On an unrelated note, the co-creator of 'Slay the Spire' seems to like 'Inscryption' and 'Wildfrost'. I myself enjoyed 'Inscryption', but I haven't tried 'Wildfrost'.
I don't know why Yelp is so much worse than Google Maps (or OpenTable or TripAdvisor!) but I agree that it's terrible. For some reason, there are lots of people who rate their local Taco Bell, Burger King, or adequate-but-not-great-local-bar-that-serves-pizza-and-wings as 5. (I understand the last one - it's a place they're fond of.)
Presumably, the next killer advance would be some kind of smart search which figured out if people who have similar tastes and profiles to me liked this restaurant. If privacy rules let you, somebody like Google could incorporate a lot of data beyond reviews.
Google Maps and TripAdvisor filters are good enough to suggest which places I should read the reviews for, and the reviews are pretty helpful in figuring out "would I like this place," but Yelp is almost negative value. (Google is particularly good for "I am at a hotel in a neighborhood I don't know. Is there a place in walking distance that I would like significantly more than the hotel restaurant that is open right now?")
I've found this to be geo-specific. I lived in NYC from 2021-2023, and the Google Maps ratings were extremely discerning. Any restaurant over 4.5 with hundreds ratings was guaranteed to be good. Yelp was worse.
In SF, it's the opposite. Google Maps feels inflated and random. Yelp is a better source there.
Maybe there's some path dependence where platforms first take off, and then an aggregation effect there.
Overseas right now and "ratings are worthless here" is a common expat trope. Curated lists by magazines or bloggers or Michelin plus word of mouth basically all you can do.
Yelp is useful at helping you avoid "don't go here, tourist trap" restaurants.
When we travel to Europe and we are standing outside an open restaurant, we use both TripAdvisor and Yelp to check the ratings. Accurate results for a good experience about 95% of the time.
Negative selection, especially at the low end, is an easier job, and indeed basically any review system will get this right if you actually read the details. So I can buy that this works.
I don't know why this is an unsolved problem. We have the technology for personalized recommendation engines, Netflix had big public competitions to refine these algorithms over a decade ago. Why has that never been tried for food? All the problems with who is rating or geographical differences or 5 star McDonald's seem transparently to be problems clustering everybody's preferences together. I'd welcome an app that said "ok try both this popular greasy spoon and this tiny Bulgarian place so we can better map your preferences" when I open it.
I live in San Francisco and eat out pretty much all the time. Zvi is correct, at least in SF: DoorDash and Yelp ratings are effectively useless; the very good restaurants are virtually indistinguishable in ratings from the merely good or even average. Google Maps does offer some signal, fortunately, whereby the "very good" classification reliably begins at 4.6 stars, and a 4.8 is quite spectacular. (This is of course relative to price – Dinosaurs Bahn Mi is a 4.6 and a mere $, while Sociale is also 4.6 but $$$.)
Of course, as in the 1990s and before, there is no substitute for personal discovery, and the average star rating on Google Maps does not always apply to an individual dish. (The Pan'terra at Pasta Supply Co. is heavenly; the meatballs are merely average.)
I’ve always wondered why Zvi eats animals. I’m still wondering…
He says a little about his own approach to morality in the New and Old EA Cause Areas section at https://thezvi.substack.com/p/book-review-going-infinite. It seems like he doesn't think of value as a fully systematizible thing, and instead aims to use consequentialism as a heuristic (possibly not the right word) for identifying tradeoffs that seem like they are worth taking. As opposed to thinking of consequentialism as a guide to tradeoffs that always ought to be made, given a certain level of confidence about a certain number of somethings.
Let's apply the utility monster thought experiment to a hypothetical superintelligence. Wouldn't a superintelligence that was capable of killing all humans probably also be capable of making so many copies of itself that the value-maximizing action would be killing off humans to a) use their atoms and b) prevent them from turning the superintelligence(s) off? In this scenario, do you find that you, too - for ultimately 'irrational' reasons - don't want to treat value as a fully systematizible thing?
If so, you might find it a little easier to see why Zvi would also opt out of trading off human interests against animal suffering.
I understand this. Zvi says he 'mostly does virtue ethics.' Is it virtuous to fund the torture and slaughter of sentient animals?
I don't mean this to be combative, but isn't the most likely reason nothing but the main: that animals are delicious, and Zvi has concluded that the benefits he receives from eating animals is more important to him than the suffering the animals undergo as a result?
Alternatively, https://www.smbc-comics.com/comic/singed
A few comments.
First of all, I don't know how Yelp is still useable. They've long since been hiding reviews (as "unreliable") an directing you to use the app, which is annoying enough that I barely use them, except if I'm really digging for information on a place.
To me, internet "ruining" things is making them popular, so the demand makes it too expensive, or ruins the quality, or the vibe. A similar thing is the case with "undiscovered" places and anything that's a good deal. These exist but not on the easily accessible public Internet.
I agree Yelp reviews do not seem helpful. I do have quite a bit of success with Google Maps ratings, but I’ve found it takes some calibration for the specific area. Some areas the baseline review is a 3.5 and a 4.5 has a good chance of being great, while other areas the baseline seems to be 4 or 4.5 and you have to be more discriminating. At the high end, the signal starts to get lost when the baseline rating is around 4.5 and I have to rely more on other methods.
So the best piece of restaurant advice I got from Tyler Cowen, in An Economist Eats Lunch, was to look for small, family owned restaurants likely subsidized by a larger business, because they're likely to be both cheap and extremely good.
His *specific* advice was to actually look for a Thai or Vietnamese place literally attached to a small motel, because that's a clear signal of that. Mine is the generalized advice I abstracted from his specific suggestion.
It served me well on a recent ~4k mile road trip across the USA, and I found numerous "grandma's recipe" style places that were indeed both cheap and excellent.
Other examples besides motel / lodging would be gas station, jerky store, butcher, winery, and farm. And I'll happily double down on "farm," if you can find a restaurant run by a farming family, go even if it's an hour round trip out of the way and the hours are terrible, because the recipes, execution, and ingredients will all be top notch in my experience.
An extreme version of this idea - When I was in high school Toyota built a factory in my hometown. A nice sushi restaurant opened up right next to it and would sell expensive sake to the Toyota execs and very reasonably priced lunch specials to anyone else. Still some of the best sushi I've had outside of Japan.
I did find that Yelp reviews were far more useful when I lived on the west coast, but that was 5-10 years ago. I don't know whether the population of users leans more hipster there, or if the service has just declined in general over that time. I would believe either or both.
There's something to the 'hipsters not tourists' point, but I don't *exactly* agree with it. Rather, one should indeed look at who is doing the rating.
I like South Asian food quite a lot, but the ratings on an Indian place in a mostly white neighbourhood/town are completely useless, whereas the rating for a place in an extremely South Asian neighbourhood (e.g. Green Street in London, Brick Lane too touristy now) are in my experience highly reliable.
Ideally the reviews on a Chinese place (assuming you're in the mood for Chinese-Chinese, not American-Chinese, noting both have their place) aren't even in English and you need to translate them.
Pub reviews are tough because ideally I want them split on food/drink/vibe, as the three are largely uncorrelated with each other and it really depends what I'm in the mood for.
Yep, there's definitely some of this. Asian restaurant ratings are by far the least reliable, many of my favorite places there don't review well at all.
Yelp is unusable as far as I can tell. Other ratings seem to select for something just beyond mediocrity. You'll see chains that are good at consistently producing a predictable, average or slightly above average rated just slightly behind what a better quality non-chain with similar pricing. A non-chain that's higher quality and higher pricing will be even closer to the chain because it gets judged more harshly by the people that are looking for good food and gets dinged for prices by people that don't appreciate quality improvements beyond that produced by chains.
Ratings in heavily tourist areas do seem to be almost worthless. A common theme for the vacation areas we live near is for a place to open up and be good, then get popular, and then either get overwhelmed by the volume and stop doing what they do well, or more typically, they take more and more shortcuts and use more and more pre-prepared foods to handle the extra volume. The second approach barely hurts the restaurants reviews for years. You will see tourists leaving those traps proudly proclaiming that the food they just ate that was basically moved off a sysco truck to a freezer, and then to an oven, griddle, grill, or fryer with no additional work necessary other than (possibly) thawing, as the best they've ever had of that type of food. Which is fine. I'm glad the combination of their expectations and good vibes from vacation produce an enjoyable experience for them. It does sort of suck that new places that are good get discovered and overwhelmed much more quickly, but I doubt the owners of those restaurants share my opinion on that though.
On an unrelated note, the co-creator of 'Slay the Spire' seems to like 'Inscryption' and 'Wildfrost'. I myself enjoyed 'Inscryption', but I haven't tried 'Wildfrost'.
Here are the links:
https://caseyyano.com/on-games-2023-565c7d2b9d63
https://caseyyano.com/on-games-2021-7617fb9e8654
I don't know why Yelp is so much worse than Google Maps (or OpenTable or TripAdvisor!) but I agree that it's terrible. For some reason, there are lots of people who rate their local Taco Bell, Burger King, or adequate-but-not-great-local-bar-that-serves-pizza-and-wings as 5. (I understand the last one - it's a place they're fond of.)
Presumably, the next killer advance would be some kind of smart search which figured out if people who have similar tastes and profiles to me liked this restaurant. If privacy rules let you, somebody like Google could incorporate a lot of data beyond reviews.
Google Maps and TripAdvisor filters are good enough to suggest which places I should read the reviews for, and the reviews are pretty helpful in figuring out "would I like this place," but Yelp is almost negative value. (Google is particularly good for "I am at a hotel in a neighborhood I don't know. Is there a place in walking distance that I would like significantly more than the hotel restaurant that is open right now?")
I've found this to be geo-specific. I lived in NYC from 2021-2023, and the Google Maps ratings were extremely discerning. Any restaurant over 4.5 with hundreds ratings was guaranteed to be good. Yelp was worse.
In SF, it's the opposite. Google Maps feels inflated and random. Yelp is a better source there.
Maybe there's some path dependence where platforms first take off, and then an aggregation effect there.
Overseas right now and "ratings are worthless here" is a common expat trope. Curated lists by magazines or bloggers or Michelin plus word of mouth basically all you can do.
Yelp is useful at helping you avoid "don't go here, tourist trap" restaurants.
When we travel to Europe and we are standing outside an open restaurant, we use both TripAdvisor and Yelp to check the ratings. Accurate results for a good experience about 95% of the time.
Negative selection, especially at the low end, is an easier job, and indeed basically any review system will get this right if you actually read the details. So I can buy that this works.
I don't know why this is an unsolved problem. We have the technology for personalized recommendation engines, Netflix had big public competitions to refine these algorithms over a decade ago. Why has that never been tried for food? All the problems with who is rating or geographical differences or 5 star McDonald's seem transparently to be problems clustering everybody's preferences together. I'd welcome an app that said "ok try both this popular greasy spoon and this tiny Bulgarian place so we can better map your preferences" when I open it.
I keep telling people to start this company or build the thing but no one does.
I might work on it in my Copious Free Time, Real Soon Now, but things have been kind of busy.
I live in San Francisco and eat out pretty much all the time. Zvi is correct, at least in SF: DoorDash and Yelp ratings are effectively useless; the very good restaurants are virtually indistinguishable in ratings from the merely good or even average. Google Maps does offer some signal, fortunately, whereby the "very good" classification reliably begins at 4.6 stars, and a 4.8 is quite spectacular. (This is of course relative to price – Dinosaurs Bahn Mi is a 4.6 and a mere $, while Sociale is also 4.6 but $$$.)
Of course, as in the 1990s and before, there is no substitute for personal discovery, and the average star rating on Google Maps does not always apply to an individual dish. (The Pan'terra at Pasta Supply Co. is heavenly; the meatballs are merely average.)