great insight, thanks.
3 Likes
Very interesting post. Caveat Iām not a data person, but Iām skeptical, respectfully, that some of the conclusions can be drawn from the underlying data (mine included). But I may be out of my depth.
From this, we donāt have an index for each inspector, their relative leniency (more on that later). Or for each restaurant, itās hard to know whether a change in grade is a change in material conditions (vs inspector leniency) without contemporaneous inspections by different inspectors. And are neighborhood comparisons apples to apples (different mix of restaurants)? Thereās also no assessment of possible bias based on cuisine, price point, or some other factor.
Curious about the data and the ability of AI to process data, I put all this into Claude. Basically it scraped the county data, analyzed by inspector ID, and did find thereās a new, seemingly harsher inspector in those neighborhoods, though the drop didnāt quite coincide with the new inspector.
Per Claude:
-āThe share of inspections earning an A went from 94% (July 2023āJune 2025) to 75% (first half of 2026). In the rest of the city it went from 96% to 93%.ā
-āCountywide, inspectors in their first two years score the same restaurants within about half a point of veterans, so the gap looks specific to this inspector. Comparing the same restaurants across inspectors, this one comes out stricter than about 94% of county inspectors. Thatās a rough estimate from 27 inspections outside these neighborhoods. The long-time inspector they replaced was on the lenient side, so some of the drop is a lenient inspector leaving.ā
But take all this with a grain of salt. Using Fable and Opus for other projects (health/healthcare and document based research), Iāve found it overinterprets data and will qualify claims when prompted to review. And Iām skeptical that a new inspector scoring the same restaurant lower than the previous inspector necessarily proves anything. Itās suggestive, but there could be other factors, including actual changes in condition of the restaurant.
*Trying to find a way to publish the whole Claude report anonymized.
3 Likes
Youāre skeptical that which conclusions can be drawn from the underlying data?
The data is the data is the data - you looked at a few additional angles - some of them quite interesting - but I donāt think thereās any disagreement or contradiction between our respective analyses.
3 Likes
Agree. Appreciate your post.
Iām skeptical that the conclusions flow from the data. For example, if a restaurant has more small violations from one inspection to the next, we canāt know whether previous inspectors saw these and let them go, missed them, or the kitchen genuinely was dirtier from one visit to the next. And even if it was the same inspector across visits, you still canāt prove those things.
From the data it seems thereās a new inspector there (this is where we disagree I think), they write more violations than the previous in that area and wrt the city in general. Maybe thatās a harsher inspector, maybe those restaurants were genuinely dirtier, maybe the inspector holds a bias. But given that the county lists inspectors, neighborhood averages donāt per se disprove that thereās not a new, harsh inspector.
Indeed. Your analysis appears to show there is a new, harsher inspector. Thereās no disagreement there.
But the disconnect between our conclusions relates to magnitude.
This new inspector, on average, scores ~4-6 points harsher. Firstbornās latest inspection resulted in 39 violation points. Their previous graded inspection in June 2025 resulted in 3 violation points (97/100). Take 4-6 points off that score and Firstborn would still receive an A.
So there is a harsher inspector, but the data does not support pinning all or even most of the āblameā on that inspector.
2 Likes
This seems like a reasonable exception to the rule, since itās interpreting data and not expressing an opinion about food, and you detailed some of the modelās weaknesses, but the rule here is not to post AI output.
3 Likes
I get what youāre saying, but both my and @Loaxley 's post have graphs generated by AI. All AI writing was labeled in my post (assumed the graphs didnāt need labeling as theyāre clearly AI).
Is there a forum AI policy? Wasnāt aware, happy to comply though.
Thereās way more overt and useless AI drivel across this board that hasnāt been policed. These uses feel like the prime example of it being used as a tool rather than a pure creation mechanism so I think having rules that dillineated between the two would be massively helpful.
3 Likes
Seems like a counterproductive of waste time to try to make rules that cover every situation. Better to have simple rules and enforce them with discretion. Like I said, this was a reasonable exception to the rule, which seemed like a good opportunity to mention the rule for anyone who might not have been aware of it.
5 Likes
Hear hear! so hard to avoid the slopficiation of EVERYTHING so nice to see some lines drawn. Do the work if you really want to parse this. Itās not a tool, itās a crutch.
LLMs are useful tools for a lot of things, but, except in rare circumstances, not for talking about food.
1 Like
In no way would I be able to find the time in the day to find, download, parse, and visually present this data without using LLMs. Itās not doing anything I couldnāt do. Itās not doing anything differently than I would do it. Itās just doing it faster and generally prettier (when it comes to data). LLMs, like calculators or auto-complete, are simply tools, just as you said.
However, I donāt like using AI for most writing*, even though I am well aware that Iām not a particularly eloquent writer.
*Getting things back on topic, I did use AI to fill out a form and draft an email to LA County to request a copy of the firstborn inspection. A prime example of the only sort of writing for which I use AI.
3 Likes
Yeah, when an LLM can deal with paperwork without making mistakes, thatās great.
okay ā¦ iāll just spend 5 hours doing what OpenAI just did in five minutes with better accuracy ā¦ for the principle of the matter ā¦ because iām anti ā¦ the unfairness of it all
1 Like
Just donāt send Claude to eat at a restaurant and tell you how you liked it.
3 Likes
there are already AI bros out there trying to āsolve restaurant reservationsā so donāt act like itās not going to come for things you love just because it hasnāt affected you yet.
2 Likes
FIFY
for some hard to get restaurant reservations, these AI services are pretty much empty and kind of a scam in some cases. particularly for bookings in Japan, Iāve seen an AI company advertise as if they could get reservations at some restaurants where specifically they wonāt be able to work. some exclusive booking platforms simply do not work in a way where the AI can add anything of real value. nonetheless, they can sell users the idea of ādonāt miss outā but structurally they wonāt be able to do anything that helps the user in many cases.
but in many other cases, and yes for booking appointments/reservations, there are a lot of ways that AI can make things worse for us.
1 Like
Iām sure this gets split off but a massively discussed part of AI discourse today when it concerns reservations. A solution in search of a problem from people outside an industry who donāt know shit. A bunch of agents telling restaurants that their boss deserves a reservation more because heās a special boy and itās his birthday is the future of hospitality. But par for the course for most AI usage today, a bunch of people who want to feel like a special boy because they can be a creative while skipping the act of creativity focused on the result and not the craft.
when it comes to great productivity idea, research, data, science, medicine, etc Iām all for it. Like how posters I respect in this thread used it to hone their point of view. But when it comes to creativity or the arts or food or anything that has craft - thereās a fine line between using it to render VFX and hone your skills vs the common application of it now which is āhereās your āartāā that makes the universal dismissal of any anti AI point of view as being a Luddite or āboo hoo you think itās unfairā the most annoying thing. And to that I say, touch grass.
7 Likes