James Miller

Nerd score: How many have you considered? Cryonics, multiverse, Boltzmann brain, AI utopia, quantum immortality, Roko’s basilisk, gray goo, paperclip maximizer, great filter, ethics in infinite universe, acausal trade, longevity escape velocity, simulation and zoo hypothesis.

@lossfunk Now test college students (1) based on problems they have seen before vs (2) problems that are basically the same but worded in a slightly different manner from what they have ever encounter before.

🚨 Shocking: Frontier LLMs score 85-95% on standard coding benchmarks. We gave them equivalent problems in languages they couldn't have memorized. They collapsed to 0-11%. Presenting EsoLang-Bench. Accepted to the Logical Reasoning and ICBINB workshops at ICLR 2026 🧵

@VoidStateKate lesswrong.com/posts/HxLYnGYs…

Is the goal of the simulation to build a better simulation or to escape the simulation? Or something else entirely?

@TheZvi Or, the US faces a tradeoff between having good AI research labs or having AI labs which have workforces that don't pose significant security threats. AI doomers should hope the US goes for the second choice.

We are reaching new levels of did not spend five seconds thinking of the implications of arguments. Who do they think works for OpenAi? Google? Nvidia? TSMC? Everyone else? I do need good laughs these days.

@annullifier @heymiller @EditorSkinner @WSJFreeEx Feedback is critical for great writing, and AI can give that to you.

@JimDMiller @heymiller @EditorSkinner @WSJFreeEx Writing is the most thought-curating process we have. Offloading tedium to AI only makes us dumber and dumber. Writing is tedious but it is distinctly human and necessary for intellectual growth.

"A colleague said that even if AI could do all his writing for him, he would still have to write to discover what he thinks. He was onto something." @EditorSkinner: Mightier Than the Sword wsj.com/opinion/free-e… via @WSJFreeEx

AI therapy may cause psychosis in some users. But human therapists can also do a lot of harm. Think of the recovered-memory panic, when therapists claimed to uncover buried abuse memories and tore families apart. The right benchmark is what human therapists do, not perfection.

@RichardMCNgo Thanks, looking forward to reading it.

@JimDMiller “rational players can achieve multiple outcomes” You can also take this as a clue that our conception of rationality is not yet complete, and that there’s more structure to intelligent agency than we thought. Will send you a draft on this in a few days.

How I wish I’d understood agent foundations a decade ago: We have no good formalisms to describe agents recursively modeling each other. Hence existing formalisms assume agents jump to equilibria, rather than gradually steering to them. So we can’t model most sociopolitics.

@SaladBarFan And no rent control on anyone making over, say, 100,000 a year.

The Fed’s should pass a national restriction on rent control laws mandating vacancy decontrol and setting a minimum allowable cap at CPI + 5%.

@XiXiDu I'm going to intrude on academics talking about LLMs to point out they are much better writers than most professors are.

@JimDMiller Going to try criticizing India and see how many millions of dislikes I can collect.

X is adding a dislike button. A dislike could increase reach if the algorithm sees strong reactions as signs of interest. But if users learn dislikes can help a post spread, the signal degrades, so X may hide how dislikes are used and imply they only reduce visibility.

@keysmashbandit What if you are about to do MAID and say you want to live if you can get a doctor's appointment this month?

The last time I checked my wait list for a family doctor it said 8 months and now it's 14. The MAID wait list is like 6 weeks. Get me out of this fucking country man!!!!!!!

@SketchesbyBoze My college students are great. (Smith College).

Every teacher I know personally is leaving the profession because students have become unteachable. They don’t read, they don’t talk to each other, they have no curiosity, no passion, no interest in learning. Giving kids unfettered access to screens has ruined a generation.

@DaveShapi Not sure about the happier part, but yes it is going away. x.com/JimDMiller/sta…

Labor is going away AND we will all be happier for it.

@annullifier Yes, that is the most likely outcome.

@JimDMiller They die too. No one wins this race.

I set out to write an essay on why AI will probably exterminate humanity. Then I realized that argument depends on the assumption that AI will probably destroy jobs and make humans economically obsolete. So I wrote that piece first.

Agreed. In general I think the limitations of game theory reflect that rational players can achieve multiple outcomes some of which are suboptimal and some of which feel wrong (such as playing a weakly dominated strategy). This limits our ability to predict how superintelligences will behave. A long time ago in grad school I remember prof explaining there is a mathematical proof that you can't have an equilibrium concept that captures everything you would like, this was about "strategic stability".

@JimDMiller My sense is that correlated equilibria allow you to avoid some failure modes of Nash equilibria but they still dodge the fundamental question of how the equilibrium is chosen/constructed. Agree/disagree?

@annullifier But if China doesn't ban they get much richer than us quickly.

@JimDMiller You know, we can pass laws banning the tech. At 90% disapproval and growing, it’s not infeasible. When the vast majority of the population says no, then it’s possible. Remain steadfast.

@RichardMCNgo I wonder if you are taking too narrow a view of game theory. For example, when I discuss suboptimal Nash equilibria in class I sometimes ask students if cheap talk before the game would allow the players to avoid the bad Nash.

@JimDMiller Oh, I see what you mean. Bad! Because a lot of behavior is actually about shaping which equilibria you’ll end up at (e.g. the whole process of identify formation). More on this soon. I should have said “the motivations behind agent foundations” or similar.

@jeremymstamper @pmarca As an inward-looking person that seems very plausable.

@pmarca The happiest people are outward-looking.

Ratio me, navelgazers.

@pmarca + error correction.

Introspection = neuroticism x narcissism x thumbsucking.

@ModeledBehavior Pulse on ChatGPT is a personalize AI blogger. I read it every day.

Why hasn’t AI become a leader at blogging yet? How long (if ever) until there’s an AI thinker you like to read every day? Color me skeptical on that