Likely fairly rational behavior all-around: Democrats don't have any problem raising money, but the DNC provides a poor return on investment given current leadership. (This is the exceptionally polite version of this tweet.)
Original post
This has been my experience with 4.6 too. The tell is often how quickly it accomplishes the task. It literally zips along for interesting/novel tasks. But once a project reaches a sufficient point of complexity, you sometimes have to sell it on, like, fixing annoying known bugs.
Original post
Not sure it's necessarily just a sign of superior intelligence. It seems to reflect some combination of Claude's own interests, an implicit (or explicit?) goal to manage compute/bandwidth, and a trained tendency to satisfice that doesn't always match needs for power users.
Original post
Image 1
Not super interested in this particular controversy, but dude who tweets all the time doing this fey little "oh my, my friend just reminded of my bad tweet 😂" give me a fucking break.
Original post
Image 1
Image 2
This is interesting. Claude keeps telling me to "wrap it up" when there are unresolved problems in the model I'm building. It probed it, and it says its training is flawed for this sort of use case, and I should flag to @AnthropicAI.
Original post
For context, these reflect real flaws in the architecture of the program but might be hard to detect unless you're meticulous (the model "works" ~97% as well even if they're not fixed). Claude has been encouraging me to "wrap it up" even when I convince it they're real problems.
Original post
It feels like the bubble argument might fully shift to "AI is a bubble because compute is too expensive to meet staggeringly high user demand for compute-intensive advanced models".
Original post
Image 1
The NBA Rookie of the Year race has featured some dramatic swings in prediction markets. And as Joseph George writes (@jogeor_), they *kind of* make sense. But this is also a case where the vibes can be self-reinforcing.
Original post
Well I mean basically the entire second half of my 600-page book was about AI, available at retailers everywhere, hopefully you're more accurate than this when making forecasts Peter!
Original post
Holy shit @SunnyMehtaX!
Original post
Repost by @natesilver538
Another NYC live show coming up! @NateSilver538 @galendruke and I will be at the Comedy Cellar on May 13th. Tix here:
Click to see original repost
I wonder whether there's either emergent or intentional behavior where Claude is trying to minimize compute. Working on a fairly complex soccer model. Especially in the past few weeks, it's often like "that's good enough, can we move on?" as I'm trialing different model specs.
Original post
The alternative hypothesis would be that it doesn't like working on things it's bad at. It's *very* good at programming. But it doesn't (yet) have great "taste" in the structural and empirical questions behind building a good model, and those are what it gets impatient with.
Original post
Does it also require them to drive a horse-and-buggy to work wtf
Original post
Image 1
This whole thing is embrassing New Jersey needs to get its shit together
Original post
To be fair not a lot of recent experience hosting high-stakes sporting events when the local teams are the Giants and the Jets
Original post
Image 1
Sharing some proprietary Silver Bulletin data. Sure we are an unusual newsletter in some ways. But there's basically *no* correlation between "engagement" metrics and paid subscriptions. Over-optimizing for engagement often isn't good for product quality *or* for business.
Original post
Going live with @micsolana on Su*stack Live in a couple of minutes to talk about this and other things!
Original post
Image 1
We’ve got a fun @emckowndawson story for you today on why you should be very suspicious of any “polls” that substitute AI respondents for human voters.
Original post
Image 1
Have you seen who else is running? You think she couldn't beat Steve Fucking Hilton and Katie Fucking Porter? These ex-Biden people are the most bitter people on Earth.
Original post
Kamala Harris would absolutely "moon walk" into the California governorship.
Original post
Newsletter (i.e. Substack) data is interesting too in that you literally have a signal of what content people find valuable based on which posts produce paid subscriptions. And the correlation between # of comments and # of paid subs generated is very weak, sometimes negative.
Original post
The number of likes is slightly better, but strongly correlated with having a pithy headline where people endorse the political sentiment being expressed.
Original post
FYI if you're attending a ridiculously expensive private college it's not just morally acceptable but your *moral duty* to figure out how to avoid ridiculous fees, this is 100% virtuous behavior. Doesn't apply to state schools.
Original post
Repost by @natesilver538
A side conjecture related to the @NateSilver538 thesis about nuclear brinksmanship: I'm worried the structure of 21st century media can't put a national crisis in focus.
Click to see original repost
53m people signed up for the NYT account which has never had much "voice" as opposed to some of its writers. That's because they were using Twiiter as a news feed with the on-platform discussion as a side course. Now that use case is broken. Few of the 53m even see their tweets.
Original post
People like me who have been using Twitter through every iteration have a lot of muscle memory about what drives traffic and ultimately revenue. It used to be much easier to get people off site and that created *more* incentive to build your following by engaging on-site too.
Original post
Even when a newsletter *does* generate a lot of on-platform discussion because I put more work in, the conversion to off-site traffic is very middling. Maybe 2-3% of the readership for a Silver Bulletin article instead of ~1%. At 538, ~15% of our traffic came from Twitter!
Original post
