Updated 0s ago · Free board, delayed at least 60 s
Nate Silver @NateSilver538 on X
Nate Silver has posted 0 times so far today, Sunday, November 22 (New York time), against a usual 1 by this hour.
- Today
- 0
- Usual by now
- 1
- Last 7 days
- 54
- Daily average
- 7.7
Below usual
Latest posts
- Quote

I dunno, but plausibly the productivity is multiplicative, not additive. Like, you need both a very smart human and a very smart computer. For *right now* the people who seem to be getting the most use out of LLMs are people who already have a lot of domain expertise etc.
The Riemann hypothesis has resisted all human mathematician effort but has now been solved by a lab employee working on it as a “hobby” on his spare time with access to internal models and extremely high compute (latter of which we can tell even based on sheer size of this proof…
Quoting Talia Ringer 🕊🪬Yeah people keep pinning Levent as a non-expert AI bro because he (1) works at an AI company, and (2) posts on this website like a bro. But Levent doing cool t… - Reply
- Reply
- Reply

@NateSilver538 The overriding impulse of election science denialists, like climate science denialists, is to treat the subject as inscrutable and mock the notion that we can conclude anything. They will nitpick individual measures without proposing alternatives.

@NateSilver538 @jonathanchait Sorry, Nate, I don't think you're Isaac Fucking Newton! You're a clever pundit with a calculator.
- RepostNate Silver reposted

The other thing is to remember that long before anyone was doing statistics on this, we had these notions. E.g., why did Roy Moore lose to Doug Jones in bright red Alabama? You don't have to be a statistician to say "if the GOP nominates a better candidate they win".
I mostly try to stay out of the WAR wars re: candidate selection. But to be clear, one side of that debate is almost ... denialist. Yes, candidate strengths exist; they can be measured (imperfectly) and predicted to some degree.
- Reply
- Reply

Idk, if you don't use the models for coding or data-oriented tasks, you might think this. And I don't think super-skilled coding = super-intelligence. But if you do, it's obvious the models are pretty fucking smart and rapidly improving.
I think AI models have probably reached a permanent plateau. Have seen little improvement in last 9 months. RSI is a more complicated problem than people think: if models have natural limits there is no RSI that can overcome them
- Reply

Idk, if you don't use the models for coding or data-oriented tasks, you might think this. And I don't think super-skilled coding = super-intelligence. But if you do, it's obvious the models are pretty fucking smart and rapidly improving.
I think AI models have probably reached a permanent plateau. Have seen little improvement in last 9 months. RSI is a more complicated problem than people think: if models have natural limits there is no RSI that can overcome them
- Quote

Idk, if you don't use the models for coding or data-oriented tasks, you might think this. And I don't think super-skilled coding = super-intelligence. But if you do, it's obvious the models are pretty fucking smart and rapidly improving.
I think AI models have probably reached a permanent plateau. Have seen little improvement in last 9 months. RSI is a more complicated problem than people think: if models have natural limits there is no RSI that can overcome them
- Quote

Stephen Miller the most obvious "anonymous" source since Jared and Ivanka.
A White House official tells me that one of Stephen Miller’s first tasks upon becoming chief of staff (with Trump’s direction) was to search for a replacement for national cyber director Sean Cairncross. The official says Cairncross “did not leave, he was fired,” noting that the…
- Reply

Anyway, as you can probably tell, I had fun writing today's newsletter. https://www.natesilver.net/p/the-white-house-needs-a-ground-stop
- Quote

New analogy for what the last two years of Trump’s White House might look like just dropped.
🎧Latest Still Counting🎧 We talked the politics of AI (are tech CEOs actually really bad at it?) and @NateSilver538 spoiled the ending of Demi Moore's "The Substance." https://podcasts.apple.com/us/podcast/tech-is-breaking-is-anyone-in-charge/id6804845033?i=1000802820289
- RepostNate Silver reposted

🎧Latest Still Counting🎧 We talked the politics of AI (are tech CEOs actually really bad at it?) and @NateSilver538 spoiled the ending of Demi Moore's "The Substance." https://podcasts.apple.com/us/podcast/tech-is-breaking-is-anyone-in-charge/id6804845033?i=1000802820289
- Reply

Yes totally. The finding about attendance is based on an extremely elaborate college hoops model (COOPER) that we've spent 500+ hours on over the years. There are all kinds of controls. I don't feel any need to prove anything by using pretentious academic prose.
actually it's basically fine to use somewhat-causal language for not-very-causal designs 1) the causal designs are massively overrated in their ability to claim right to use causal language 2) most reasonable uses of data don't require you to know causality (if ticket sales pred…
Quoting John B. HolbeinLots of people dunking on this research design. That’s fair. But I’d like to focus on something else: the language. The language two-step people doing condit…
- Quote

Yes totally. The finding about attendance is based on an extremely elaborate college hoops model (COOPER) that we've spent 500+ hours on over the years. There are all kinds of controls. I don't feel any need to prove anything by using pretentious academic prose.
actually it's basically fine to use somewhat-causal language for not-very-causal designs 1) the causal designs are massively overrated in their ability to claim right to use causal language 2) most reasonable uses of data don't require you to know causality (if ticket sales pred…
Quoting John B. HolbeinLots of people dunking on this research design. That’s fair. But I’d like to focus on something else: the language. The language two-step people doing condit… - RepostNate Silver reposted

Replying to @lxeagle17
Building this was not easy and took multiple 4AM nights, so I really hope people find it useful! For researchers, academics, and nerds: we have a full, downloadable candidate quality database for all federal elections between 2014 and 2026, viewable here. https://split-ticket.org/full-wins-above-replacement-war-database/
- Reply

I think what's noteworthy about Silicon Valley CEOs, not just in AI, is the degree to which they're skating behind the puck. Like it shouldn't have been *that* hard to imagine 2024 being the peak for MAGA rather than a takeoff for it.
.@galendruke and @NateSilver538 on whether AI CEOs are good at politics.

I'm not going to claim I did a *fantastic* job of that, but I think I did better than Silicon Valley. Like, this article from a week after the 2024 election holds up pretty well, viewing it as analogous to 2004 (somewhat implying the backlash in 2006/08) https://www.natesilver.net/p/its-2004-all-over-again
- Reply

I think what's noteworthy about Silicon Valley CEOs, not just in AI, is the degree to which they're skating behind the puck. Like it shouldn't have been *that* hard to imagine 2024 being the peak for MAGA rather than a takeoff for it.
.@galendruke and @NateSilver538 on whether AI CEOs are good at politics.

I'm not going to claim I did a *fantastic* job of that, but I think I did better than Silicon Valley. Like, this article from a week after the 2024 election holds up pretty well, viewing it as analogous to 2004 (somewhat implying the backlash in 2006/08) https://www.natesilver.net/p/its-2004-all-over-again
- Quote

I think what's noteworthy about Silicon Valley CEOs, not just in AI, is the degree to which they're skating behind the puck. Like it shouldn't have been *that* hard to imagine 2024 being the peak for MAGA rather than a takeoff for it.
.@galendruke and @NateSilver538 on whether AI CEOs are good at politics.
- Quote

Doesn't make a lot of sense. It's not an easy race. Vance underperformed in his one Senate win and Brown is an overperformer. And we'll see but 2028 could be another blue or blue-ish year.
Politico: Some Republicans in Ohio are hoping JD Vance runs for Ohio’s Senate in 2028 if he decides not to run for president “JD Vance would be a dream candidate,” said one Ohio Republican. Link to article: https://www.politico.com/news/2026/11/18/senate-map-2028-bernie-moren…
- Reply
- RepostNate Silver reposted

BREAKING: Democrat Mary Peltola defeated Republican Sen. Dan Sullivan in Alaska, NBC News projects, pushing her party’s Senate majority to 56 seats after a stunning wave election. https://nbcnews.to/4iOoUkT
- Quote

The beef is that Sam Haselby is antisemitic.
no idea what this beef is about but calling someone "downwardly mobile" is such a cartoonishly elitist way to criticize someone. just say you think he's a dirty poor, nate https://x.com/NateSilver538/status/2122509913670701263?a=CEua0bcqNMEZNKHqiuQ&s=20
Quoting Nate SilverIt's a robust competition, but nobody on this site exudes more downwardly mobile public intellectual energy than you. And you have a very weird thing about Jew… - Post

The one state where Democrats didn’t have a great night on Nov. 3 was Florida. But they did have a pretty good night once you dig under the hood. Here’s Eli on a night of near-misses and where it stands in the pecking order. https://www.natesilver.net/p/did-democrats-blow-it-in-florida
- RepostNate Silver reposted

Replying to @NateSilver538
Speaking as a conservative, I'll just say this: Trump could have realized that building ASI is the least conservative, most radical, most destabilizing thing humanity has ever tried to do. He could have acted on that understanding and signed an executive order pausing ASI development. Instead, he simped for Silicon Valley, cozied up to Big Tech, betrayed conservatives, and betrayed the entire human species.
- Reply

This is related btw. In my experience, for coding-related tasks, the models are doing things extremely quickly as compared to even a few months ago. https://x.com/NateSilver538/status/2122504930082669018?a=735wepw_Mz5xuNDwYis&s=20
The rate of progress for the AI models is just so much clearer for programming-related tasks than for other things. Like even as compared to when we wrapped up work on FLIPR in July/August, the models are notably faster/better. Whereas progress seems kind of mid on other stuff.
About this tracker
This page follows Nate Silver (@NateSilver538) on X, one of the accounts on TweetStream's live board of posts that move markets, in its Politics group. New posts are listed newest first with their catch speed: the time from the post's own timestamp to TweetStream's first receipt. TweetStream catches X posts in typically 166 ms.
The count includes what shows on the account's main feed: main posts, quote posts, reposts, and replies that continue its own thread. Other replies are left out. Polymarket's weekly post-count rules describe the same kinds of post, but this is not their settlement figure: each market settles on the tracker its rules name. "Today" runs from midnight New York time. "Usual by now" is the average over the previous 7 days, scaled to the share of today that has passed, and the pace label says whether today is above or below it. A post deleted after TweetStream caught it stays in the count. Posts appear here at least 60 seconds after TweetStream receives them.
More in Politics: President Trump, Donald Trump, JD Vance, Marco Rubio, White House Press Office, Speaker Mike Johnson.



