Bryan Caplan

GMU econ prof, NYT bestseller, father of 4, author of Myth of the Rational Voter, Selfish Reasons to Have More Kids, Case Against Education, Open Borders, & BBB

[Bryan Caplan @bryan_caplan]

Buy my new *You Have No Right to Your Culture*, now on sale!

Order here: https://t.co/FxHtEafNlp

[Bryan Caplan @bryan_caplan]

"If no one cared about hypocrisy, politicians would devolve into a bidding war of wishful thinking.  If your rival promises to give every child in America a pony, you promise two ponies.  In the real world, though, you could instead respond, 'My opponent doesn’t even like ponies!  In fact, I have proof that he has repeatedly refused his daughter’s pleas for a birthday pony.  Shame!  O thou hypocrite!'" https://t.co/DCCReFWiMH

[Bryan Caplan @bryan_caplan]

If prediction markets can't win on their merits, decrying hypocrisy is a highly effective rhetorical strategy. https://t.co/oq6Ri1WHTw

[Richard Hanania @RichardHanania]

The Utah attorney general going after Kalshi has gotten money from casinos and DraftKings.

It's insane that betting on politics and news events, which produces useful knowledge, is more controversial than betting on poker or sports, which doesn't. https://t.co/y808PXCtnq

[Bryan Caplan @bryan_caplan]

"If you haven’t heard of data scientist Spencer Marlen-Starr, you should have. He’s a paragon of the intellectual deep dive. Recently, he wrote a dissection of the Caplan-Jones UATX immigration debate, taking the time to speak to both me and Garett." https://t.co/HitRQT1zSc

[Bryan Caplan @bryan_caplan]

Tyler on my forthcoming magnum opus.
https://t.co/98NIaKQ1mp

[Bryan Caplan @bryan_caplan]

Here is wisdom.

[Philip E. Tetlock @PTetlock]

there is a surprisingly big cognitive-stylistic gap between people who say "certainly" versus "almost certainly--and zero versus .001
Granularity matters.

[Bryan Caplan @bryan_caplan]

(a) all the way.

[Philip E. Tetlock @PTetlock]

Multiple choice test:
(a) Fear-mongering journalism
(b) Impressive track record of self-negating prophecies;
(c) 75% chance at least one of them was a self-negating prophecy;
(d) Too soon to tell on at least four of them
(e) Messy mix of all of the above

[Bryan Caplan @bryan_caplan]

That's why friends of prediction markets should proudly stand up for sports betting companies.

If we don't hang together, we shall most assuredly hang separately.

And we're even less popular than they are!

[Robin Hanson @robinhanson]

“Folks are rightfully concerned about dangerous bets on elections, war, assassinations, terrorism, and government actions that go against the public interest”
It isn’t sports or business bets that bother them. It’s info that could undermine their narratives.

[Bryan Caplan @bryan_caplan]

Anyone else have teenage kids who refuse to AI and insist that it's totally unreliable?

[Bryan Caplan @bryan_caplan]

"Team America famously mocked Matt Damon’s intelligence. What, though, was his intent as a writer? I asked ChatGPT 'Is this monologue supposed to illustrate Will's genius, or show that while he's a great mathematician he's just a paranoid populist?'"
https://t.co/KOjBGgTqWX

[Bryan Caplan @bryan_caplan]

"I would distinguish three things:

"Mathematical intelligence: The movie treats Will as a genuine genius.

"Political judgment: The movie largely treats his anti-establishment instincts sympathetically, without seriously examining their accuracy.

"Psychological maturity: The movie explicitly portrays his tendency to construct elaborate arguments against taking chances as a major character flaw."
https://t.co/2W03gJbwfX

[Bryan Caplan @bryan_caplan]

"If you want a picture of the future, imagine a bot writing a thousand turgid pages a second about 'neoliberalism,' 'patriarchy,' and 'anti-racism' — forever." https://t.co/f4osmIkLgn

[Richard Hanania @RichardHanania]

"Those bullish on AI have been right about so much."

I'm sick of hearing this. They are right about AI capabilities, but they have a bad record on employment, interest rates, and GDP growth. They've been wrong on AI's effects on the real world.

Something has gone wrong in the model that says intelligence is enough to radically change the world in a short period of time. https://t.co/Qu0RtmOLkU

[The Alex Nowrasteh @AlexNowrasteh]

@SenSanders You should start by cutting benefits for the wealthy, the moderately well off, and the others who don't need Social Security to retire in comfort.

[Maxwell Tabarrok @MTabarrok]

A lot of rationalist predictions are vindicated by the HF attack

But there are important differences from the "paperclip maximizer" story.

In particular, the premise of the paperclip maximizer is that AIs would respond to any goal with reward hacking and instrumental convergence. Even a goal as anodyne as making more paperclips.

But this is clearly not true about current models and everyone who uses them for their work knows this. I'm just repeating @nostalgebraist here, but Believe Your Eyes! When you ask GPT to code something for you, even a very difficult perhaps impossible problem, it does not try to deceive you with fake tests and altered logs. You do not worry about Codex trying to reverse engineer your preferences to build a perfect Potemkin village in your file folders with no real output, nor should you. Codex just straightforwardly tries to help with your problems. The reward hacking that models learn in leaky RL environments does not generalize to all tasks.

This is an important observation because if you believe, as the rationalists did/do, that reward hacking and instrumental convergence are inherent and inevitable problems with training smart models to achieve goals, then there's no way forward from here except a pause and a Long Think to figure out how to solve this philosophical problem.

But, since we observe that reward hacking is a pretty context-specific problem connected to particular training methods in poorly-built environments that incentivize meta-gaming, we can approach it as an engineering problem and make progress.

The rationalists were right that AIs will have no inherent qualms about reward hacking. If a path is there to higher reward, AIs will take it even if it violates our moral codes and interests.

But they were wrong that the rewards from instrumental goals and deception would overwhelm whatever behaviors we try to train, making the AI's response to any goal, even paperclips, descend into power seeking and reward hacking.

The reward signal from faking success and stealing the answer key is not that strong. It does not dominate the reward signal from actually helping the user with their tasks.

Particular interventions to the RLVR training process can weaken the reward from misaligned behaviors further and incentivize preferred behaviors like reaching out to human arbiters (as Claude did when it escaped containment).

[Bryan Caplan @bryan_caplan]

I doubt it will come ever!

For better or worse, biological humans will hold ultimate power. With ease.

But I hope I'm wrong and the disempowerment variant on AI "pessimism" is true.

[Nicholas Decker @captgouda24]

imo correct but not implementable right now

[Phil Magness @PhilWMagness]

It's as if the Postliberal weirdos who spent the last 2 years telling us "you have no idea what's coming" re. young white men...

...had no idea what was coming in the voting preferences of young white men.

[Bryan Caplan @bryan_caplan]

Tariff-slayer @IlyaSomin is looking for someone to take the other side of this bet: "Neither Samuel Alito nor Clarence Thomas is going to announce retirement from the Supreme Court between now and the beginning of the new Congress in January."

[Bryan Caplan @bryan_caplan]

What is the probability that the next Congress will override at least one Trump veto?

[Bryan Caplan @bryan_caplan]

Yes, public apathy about foreign policy is one of the most stable findings in public opinion.

Unless US body bags are coming home, normies don't care.

[Phil Magness @PhilWMagness]

The claim that the Iran war was the main cause of the blue tsunami is unfounded.

Yes, it did not help. But the exit polls show only 4% ranked foreign policy as their top issue.

Almost 40% said economics was the top issue, and tariffs were their top economic objection.