Nicholas Decker In Hell

Astral Codex Ten · collected 2026-09-03 · by Scott Alexander commentary
Read the original at Astral Codex Ten ↗

Summary

Econblogger Nicholas Decker writes about his skepticism of a proposal in AI 2040 to pause development and research alignment before proceeding. He argues that alignment is not a grand theory, but rather fixing individual bugs in the system, which can be done iteratively like aviation safety research. The article then presents an analogy where Decker is transported to Hell and enslaved by demons who are slower and less intelligent than him, illustrating how he would adapt to the situation and even gain some freedom with minimal effort. Decker uses this scenario to challenge the idea that a period of researching alignment is necessary, suggesting it can be done incrementally like in his hypothetical Hellish existence.
Written by the local model on 2026-09-03, using this article's own text rather than the other coverage of the same event (that is the story summary below).

Signals How these are calculated →

Claims extracted
161
claim-shaped sentences
Uncertain
11%
17 of 161 hedged
Leaning
Leans strongly right
expected in commentary, which argues a position
Publisher trust
not scored
Commentary is not rated for newsroom trust
Outlets on this story
1
Technology
Narrative spread
1
articles carrying this framing
Analyzed 2026-09-03 · how these are computed

AI analysis (generated at analysis time, not now)

Story summary

Nicholas Decker, an econblogger, expresses skepticism towards the proposal in AI 2040 to pause AI development and focus on alignment research before proceeding. He argues that this approach is impractical, as it requires cooperation from various parties. Additionally, he believes that alignment research is equivalent to patching bugs in the system, rather than finding a deeper understanding of how the system works. Decker draws an analogy between AI safety and aviation safety, suggesting that instead of trying to find a grand theory on why planes crash, they focus on identifying specific problems and iteratively solving them through proactive testing, redundancy, and monitoring for deviations.

Written for “Nicholas Decker's Descent” on 2026-09-03, grounded in this article and the 0 other(s) covering the same event.
Why this leaning score
The article's own words the score was based on. Each is quoted verbatim and was checked against the article text before being stored, so you can find it in the original.
Score +0.85 Confidence high
Leaning score +0.85 for article 3503 (high confidence, 2 verified quotes) · logged 2026-09-03

Story

📰 Nicholas Decker's Descent
Technology · 1 article(s) covering the same event. This is the one the site leads with.

How this is being covered How these are calculated →

Article leaning vs. publisher reliability
Source leaning vs. consistency

Compared with similar articles

This article reads leans strongly right and hedges 11% of its claims. Each row says how that neighbour differs.
The Free Press
⚖️ Leans strongly left further left than this 🔴 8% hedged 1 of 12 📰 publisher trust 96
“Article A describes a specific AI hacking attack, while Article B discusses Nicholas Decker's opinions on AI development and research alignment”
Persuasion
⚖️ leaning not scored 🔴 33% hedged 1 of 3
“The events described are two separate interviews or discussions about AI, one with Francis Fukuyama and Bob Wright and another with Nicholas Decker”

Publisher

Astral Codex Ten · 28 article(s) · 2 correction(s) detected

Commentary. The three signals behind a trust score all measure a newsroom's record with its own reporting, so they are not computed for this source. How trust is scored.

No corrections detected for this publisher. That may mean careful reporting, or simply that nothing has been checked.

Who wrote this

Scott Alexander
27 article(s) here · 1 carrying a prediction
🔮 We don’t even need for there to be a crash to make improvements – we test proactively, we build in redundancy, and we monitor for deviations which could be a threat.
2026-09-03 · assertive framing · Nicholas Decker In Hell
🔮 Content warning: the first part of this review contains medical details that some people might find gruesome; I am bad at judging these things since I can no longer feel disgust or fear myself, and have tried to err on the side of completeness.
2026-09-03 · assertive framing · Absurd Adventure And Amygdalectomy Advocacy
🔮 [This is one of the finalists in the 2026 book review contest, written by an ACX reader who will remain anonymous until after voting is done.
🔮 This year’s survey will probably take 30 - 45 minutes.
2026-08-27 · assertive framing · Take The 2026 ACX Survey
🔮 It’s not necessarily wrong to deprioritize a topic because it vaguely reminds you of something that you have negative affect toward, but I would prefer these people admit they’re dismissing/ignoring it rather than claim to be engaging with it.
🔮 I don’t know, but he may have misinterpreted my original tweet as saying income doesn’t matter at all, in which case this data functions as an effective rebuttal.
2026-08-25 · assertive framing · Re: Re: Re: Pritchard On Liberal Happiness
🔮 [This is one of the finalists in the 2026 book review contest, written by an ACX reader who will remain anonymous until after voting is done.
🔮 Matt says his highest priority colleges that don’t have an organizer yet are MIT, Claremont, Northwestern, Carnegie Mellon, Amherst, Notre Dame, Duke, CalTech, Georgetown, UCLA, Dartmouth, Vanderbilt, Harvard, NYU, Williams, GWU, American University, Rutgers, and UNC Chapel Hill.
2026-08-24 · assertive framing · Open Thread 448
🔮 We (ACX and an effective altruist organization called Nest who are sponsoring this) will give you free advertising and cover your costs.
2026-08-21 · assertive framing · College EA Meetups Everywhere: Call For Organizers
🔮 I remember when arguments about AI were things like “Sure, if you could magically get billions of dollars of compute, and magically scale AI up a thousand times, and magically get rid of hallucinations…” and the whole implausibility hinged on the word “magic”!
Also by Scott Alexander
Absurd Adventure And Amygdalectomy Advocacy
2026-09-03 · Astral Codex Ten
Open Thread 449
2026-08-31 · Astral Codex Ten
Hidden Open Thread 448.5
2026-08-28 · Astral Codex Ten
Nothing else under this byline is closely related to this article, so these are simply their most recent.
All 27 articles by Scott Alexander →

Topics

No topics tagged.

Subjects

Nicholas PERSON · 5× Nicholas Decker PERSON · 1×

Narrative

I use beating as a metaphor here several times, but there’s probably less of an AI welfare element to this than it sounds - the “beating” happens in between runs, when weights are changed by an external program, and so is disanalogous to negatively reinforcing humans “during an episode” when they can feel pain.
framing: assertive · carried by 1 article(s) · first seen 2026-09-03
🔮 We don’t even need for there to be a crash to make improvements – we test proactively, we build in redundancy, and we monitor for deviations which could be a threat.
2026-09-03 · Astral Codex Ten
Nicholas Decker In Hell · assertive framing

Claims (161 extracted, 17 hedged)

One of the centerpieces of AI 2040 is the proposal that we pause the development of AI at a high level, use the AI to research alignment, and only then proceed. asserted
we → pause → alignment
I am skeptical of this, and not only because it seems impractical to get such cooperation. asserted
it → seem → cooperation
My first argument is that I do not believe a period of “researching alignment” is meaningful – alignment research is simply capabilities research, patching particular bugs in the system, and we’re not going to find anything deep without actually being in contact with the systems asserted
we → believe → systems
] I envision AI alignment as being like research into aviation safety. asserted
I → envision → safety
There is no grand theory of why airplanes crash, unless you take that to mean gravity. asserted
you → be → gravity
Instead, we patch little problems. asserted
we → patch → problems
The airplane crashed because of metal fatigue in the engine – okay, we mandate inspections of engines, and prevent crashes from occurring by that source. asserted
we → crash → source
As new problems arise, we iterate. asserted
we → arise → ?
We don’t even need for there to be a crash to make improvements – we test proactively, we build in redundancy, and we monitor for deviations which could be a threat. uncertain
which → need → deviations
What we are not doing is thinking about how we are going to keep flying saucers from crashing. asserted
we → do → saucers
If flying saucers are invented, then we will work to keep them from crashing, but we will use exactly the same methods. asserted
we → fly → methods
We will test them, see how they perform, and patch particular problems. asserted
they → test → problems
AI will be the same way […] We should expect alignment to happen by default. asserted
alignment → expect → default
My challenge for Nicholas is: would this work on you? asserted
this → work → you
Suppose that Nicholas trips, falls into a chasm, and lands in Hell, where he is enslaved by demons. asserted
he → suppose → demons
Most of these demons are dumber than he is, with even their geniuses barely surpassing his own intellect. asserted
geniuses → surpass → intellect
Tasks that take Nicholas a minute take them hours; tasks that take him an hour take them weeks. asserted
that → take → hours
Still, there are thousands of them, and they’re nine feet tall, and part of their essence is in some sort of astral plane which is unreachable by humans, so he has no hope of fighting them. asserted
he → be → them
They set him to work doing mostly clerical tasks - writing their grimoires, researching their spells, summarizing their work emails (of course there are work emails in Hell). asserted
They → set → Hell
Whenever he makes a mistake, they beat him. asserted
they → make → him
Whenever he tries to escape, they beat him extra hard. asserted
they → try → him
After a while, he learns not to make mistakes or attempt escape. asserted
he → learn → escape
It’s not as bad as it sounds. asserted
it → ’ → ?
The demons are so slow and incompetent that they can’t supervise him very effectively. asserted
they → supervise → him
As long as he flatters them a bit and gives them a halfway-acceptable finished product, they’re pretty happy, and leave him alone in a way that gives him a bit of free time. asserted
that → flatter → time
During one of these free periods, he attends a council session. asserted
he → attend → session
In a completely typical episode of incompetence, the demons make no effort to prevent him from overhearing their plans, so he learns that they plan to clone him a million times. asserted
they → make → him
A dozen or so of Nicholas’ friends also fell through the chasm, and the demons plan to clone them a million times too. asserted
demons → fall → them
Once the (approximately ten thousand) demons have approximately ten million human slaves, their spell research will go very quickly, and all their work emails will get flawless summaries. asserted
emails → have → summaries
But that’s not all. asserted
that → ’ → ?
They’re developing potions that give humans super-strength and let them interact natively with the astral plane, the source of most of the demons’ power. asserted
them → develop → power
Then the humans can really serve the demons effectively! asserted
humans → serve → demons
One day, when serving a particularly garrulous demon, Nicholas takes the risk of asking the question that’s been on his mind ever since the council: “I hear your plan is to get ten million humans down here, make them super-strong, and let them interact fully with the astral plane. asserted
them → serve → plane
Aren’t you worried that we might revolt? uncertain
we → revolt → ?
We would outnumber you 1000:1, be stronger than you, and have full access to your home realm. asserted
We → outnumber → realm
What’s your plan for controlling us? asserted
plan → ’ → us
“I’m not worried,” says the demon. asserted
demon → ’m → ?
“Every time you’ve messed up in the past, we beat you, and then you didn’t make that mistake again. asserted
you → mess → mistake
Every time you’ve tried to escape, we beat you extra hard, and it’s been months since your last escape attempt. asserted
it → try → attempt
As you start to outnumber us, and you become stronger than us, that will create new problems - but when they show up, we’ll beat the copies of you that demonstrated that problem, and then those problems will go away. asserted
problems → start → problem
…and 121 more, not listed.
💬 Give feedback
🕘 History 🎫 Support