Pinned Tweet
My wife made this video to celebrate our engagement a few months ago and it's my favorite thing: youtube.com/watch?v=HWkE3g0w…
Reihan Salam retweeted
What on earth...?
New: NYC will pay the teachers’ union $8.7 million for “building positive nurturing environments” during the COVID pandemic in 2020-2022 under a RETROACTIVE contract approved last night by the Panel for Educational Policy. The DOE refused to specify what labor the UFT provided, or explain why the city is funding it SIX YEARS LATER.
Agenda item #3:
Reihan Salam retweeted
Nearly 3 in 4 NYC district students live in poverty, and 61% are black or Hispanic. Even if the city perfectly integrated its schools, the typical school would be about three-quarters low-income. Better instruction should be the focus. city-journal.org/article/new…
Reihan Salam retweeted
A lot of people were suprised by our putting community violence intervention at the top of this list.
They were right to be. In my latest @CityJournal SS, I explain that it's just not clear how reliably CVI actually reduces violence.
Link 👇
Why is homicide declining so rapidly?
A few weeks ago, I was part of a group of four dozen crime researchers convened by @CouncilonCJ to try to answer that question.
We rated nearly 50 factors based on how likely they were to contribute to the drop.
While I didn't necessarily agree with everyone's conclusions—I would describe CVI as "limited" at best, for example—it was a stimulating exercise, and I think useful for anyone trying to understand the drop.
Read the full report here: counciloncj.org/whats-drivin…
Reihan Salam retweeted
This attitude is consistent with what made @jaredpolis's generations of "Blueprint" Democrats compelling to CO's heterodox majority unaffiliated bloc.
But that coalition may be coming apart.
More below.
I’m not scared of anyone. I just care deeply about our state and don’t want the socialists to destroy it. In a Colorado for all we don’t think less of anyone because of their net worth. Whether you have no money or are millionaire, we love you in Colorado and want you thrive. ❤️
"Thanks to socialism, the average Zimbabwean became a trillionaire.”
Reihan Salam retweeted
As @tracyalloway writes in today's Odd Lots newsletter... The logical endpoint is my agent battles some company’s algorithm until they reach the same equilibrium we would reached before AI, except we both spent a lot on tokens and the whole thing is value destroying.
Reihan Salam retweeted
Striking detail in the big NYT piece on Taiwanese families adjusting to working at the huge new TSMC chip plant in Arizona: their surprise at the lack of rigor in US schools.
Reihan Salam retweeted
The active chemicals in Kratom, aka "gas station heroin", act on the same brain receptors as opioids. 7-OH (7-hydroxymitragynine) can be 13x stronger than morphine.
Federal data connect Kratom to more than 5,200 overdose deaths between 2020 and 2024. One study found dependency in more than half of regular users.
Kratom is still easily and cheaply available to buy.
Kratom, much like marijuana, cannot be safely regulated, no matter what the industry claims, @KevinSabet writes.
city-journal.org/article/kra…
Reihan Salam retweeted
This is the type of thing that potentially gets new laws passed.
While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected.
Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions.
Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent.
But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models?
We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem.
Read the full report: cohere.com/blog/concerning-b…
Reihan Salam retweeted
Good question.
Lax prosecution: Cops in the 1st and 5th Precincts made 908 arrests for trademark counterfeiting in 2025, and more than 92% of those arrests were booked as felonies. In almost all the cases, the felonies disappeared: out of 696 total counterfeiting cases in 2025, just two were arraigned as felonies, while 694 were arraigned on misdemeanors. The misdemeanor cases rarely produce serious consequences.
Inconsistent policing: Illegal vending is a low priority compared with violent crime and other demands on police manpower. Police flood Canal Street, clear the vendors, and make arrests. Then officers get reassigned elsewhere.
Lastly, illegal vendor arrests generate an unusually high number of reports to the Civilian Complaint Review Board. A retired cop told @CityJournal the tactic is often used by third parties who may or may not have ties to the vendors themselves. Once the cops get a certain amount of complaints against them, they’re benched and not allowed to leave the precinct.
A sincere question about NYC:
Why does the city allow people to set up shops on the sidewalk selling fake goods? It looks ugly, effectively steals from the shop owner's they set up in front of, doesn't generate tax, blocks walking, and frankly the people selling these good seem like a pretty bad bunch overall.
I could fix the problem in a week easily. But the city just lets it happen. Why?
Reihan Salam retweeted
Earlier this month, @NYCCouncil held a marathon hearing on AI. The snippet below is from the committee report prepared for that hearing, and shows the limits of the Council's research ability in this area.
Footnote [23] goes to a page that cites research focused on 2023-era models, with the most recent research completed in 2025.
Footnote 24, despite the text that precedes it, links to a study that was completed in October 2025, and minimally updated in March 2026 (but the underlying conclusions from October 2025, which are cited here, were not changed).
This committee report is filled with these kinds of errors, among many other kinds. I'm working on a longer post about it.
I'm not sure what to call this phenomenon. The research was done by someone who is simply not familiar enough with the domain to avoid basic errors, but they were careful to get brand name citations from MIT and academic journals. But citing hallucination statistics from October 2025, or derived from 2023-era models, is a huge error. The field just doesn't resemble that time at all right now, brand-name citations or not.
This is a kind of "highbrow misinformation." Yes, it has academic citations and such, but actually reading them shows that they aren't that valuable. One would actually be better off with good Substacks and X posts in this case.
This outdated view of hallucination also prevents people from learning: (1) how much of a problem it isn't these days, and (2) when it remains, how to use agents and other tools for productive fact-checking (see @StevenDenney86's piece here: popularbydesign.org/p/how-to…, cc: @akoustov)
Reihan Salam retweeted
One of the distinct benefits of using contractors for last-mile deliveries is that it makes change a lot faster. Companies striking the right balance between efficiency and safety are rewarded; those that aren't give way to companies that strike a better balance.
The city itself knows this, because its own agencies give out millions of dollars in contracts every year.
The so-called "Delivery Protection Act" would put a stop to that—with no evidence it will improve safety—simply because of an ideological prejudice against a big company whose services millions of New Yorkers value.
When a worker wears an Amazon vest, drives an Amazon van and delivers on an Amazon-controlled route, but Amazon denies they work for the company — that's exploitation.
When we see that traffic and injury-causing crashes spike on the streets around a warehouse — that's exploitation. And it makes our streets less safe.
We stand with working people, not corporations that profit by denying them their rights. We stand with @Teamsters and for safe streets in New York City.
We will pass the Delivery Protection Act.
Reihan Salam retweeted
If you think Midtown Manhattan in the afternoon is much, much sketchier than it used to be, you're right thebiggerapple.manhattan.ins…
Reihan Salam retweeted
Student protests in France turned into citywide riots. How?
Teens don't invent violence out of nowhere. It's learned behavior.
Consequences matter. @DrJenniferWeber ⬇️
Reihan Salam retweeted
How do we best balance security needs for AI labs with broader economic goals of containing Chinese open source models and diffusing AI?
With @fuller_brandon and @Sean_Speer we argue for antitrust exemptions paired with forced open weight releases
manhattan.institute/article/…
Reihan Salam retweeted
Read here: thebiggerapple.manhattan.ins…
