Alexander Gietelink Oldenziel

An­nounc­ing ILIAD 2026

• Alexander Gietelink Oldenziel 31 Mar 2026 16:44 UTC

  4 points

  0

  in reply to: niplav’s comment on: Take note of how bright­ness makes you feel

  Niplav, true heir of gwern

• Alexander Gietelink Oldenziel 29 Mar 2026 22:48 UTC

  4 points

  −2

  in reply to: Adam Scholl’s comment on: adam_scholl’s Shortform

  I strongly agree. The whole focus on banning AI data centers seems somewhat myopic & misguided

• Alexander Gietelink Oldenziel 28 Mar 2026 8:01 UTC

  9 points

  0

  on: The AIXI per­spec­tive on AI Safety

  As a fellow fan of AIXI, and more generally idealized models of agency and the alignment problem I’m happy to see this comparison. Imho AIXI- and AIXI-adjacent thinking is still a tiny minority, even for more mathematically inclined people in alignment. That’s a shame.

  I like your comparison of AIXI with other theory frameworks. People should make inside-view comparisons like this more often.

  That said- my main objection to AIXI for alignment is that it is blind to inner alignment. The model of AIXI already assumes you encode the agent with whatever utility function you want. That limits it application to alignment directly.

  I am most excited for modern learning theory & RL theory to meet AIXI & agent foundations. I think, at least mathematically, this is where the alignment problem fundamentally lies.

• Alexander Gietelink Oldenziel 27 Mar 2026 15:16 UTC

  2 points

  0

  in reply to: leogao’s comment on: leogao’s Shortform

  

  always had you as von stauffenberg in my head :)

• Alexander Gietelink Oldenziel 24 Mar 2026 8:37 UTC

  2 points

  0

  in reply to: Anastasia Wei’s comment on: An­nounce­ment: Iliad In­ten­sive + Iliad Fellowship

  Yes sufficient! Claude can fill in the rest =)

• Alexander Gietelink Oldenziel 22 Mar 2026 7:41 UTC

  10 points

  6

  on: China Derange­ment Syndrome

  How China fights its bordering neighbour.

• Alexander Gietelink Oldenziel 14 Mar 2026 15:24 UTC

  LW: 2 AF: 1

  0

  AF

  on: Steel­man­ning heuris­tic arguments

  >>> Basically, a very optimistic hope for understanding “why circuits behave differently from functions” is to make statements that, conditional on some “surprising” property P holding for a circuit (here “surprising” means that it occurs much more for circuits than functions), we can deduce that, with high probability, the circuit internals have some property A(C).
  
  
  My understanding of the NCP-conjecture is different [and I claim more interesting].

  If we want we can go from a surprising property Q [your P] holding for the function to some property A(C)… the canonical way is to assume some probability model and do Bayesian updating.

  In other words A(C) will be some averaged property over the Bayesian posterior. Assuming everything is discrete and finite [we work with Boolean circuit of fixed size] and a uniform prior this is all completely explicit.
  
  The more interesting thing to me is the following.
  
  Let’s say Q is a disjunction Q = Q_1 OR Q_2

  In that case, informally we could reason: well Q is the case because either Q_1 is the case or Q_2 is the case. Crucially the property A(C) holding for the underlying circuit C may depend on whether we have Q_1 or Q_2.

  Just like we can have many different proofs for the truth of the same statements more generally we can have many different heuristic arguments for the same ‘truish’ proposition. The Bayesian posterior almagates and averages over all of them.
  
  
  
  
  -------

  Random Speculations:
  
  Let me speculate on what an heuristic argument may be in this case.

  One proposal for what a heuristic argument would be is the following:

  parameterize probability distributions on the finite discrete space of Boolean circuits and Boolean functions by an exponential family [iiuc this can be done using log-linear models]. Then consider the outrageous statement Q(C) and consider those distributions that make it very likely.
  
  In other words if we parameterize our distributions by w then P(Q(C)|w) is assumed high.
  For a given C and a specific w we may ask is P(C|w) high or even equal to 1?
  In that case, that w and the associated sufficient statistics may be regarded as a reason for Q(C).
  
  of course—there are many such w potentially. We need w that are not too complicated to describe for each possible circuit C.

  Let’s look at the fiber F=F^{-1}(Q)=\{C\in Circuit| Q(C)\}.
  Now the fiber F breaks up in components F_i—intuitively there are different classes of fuctions/​circuits that can give rise to the property Q. We’re looking for simple descriptions /​ sufficient statistics w such that of a given component F_i is assigned high probability by w.
  
  How do we find these components F_i—if they even make sense? One way is to recursively break up the fiber of F of Q=Q(C) by logically decomposing Q and using subformulas to describe boolean features that may fix a sufficient statistic for above.
  

• Alexander Gietelink Oldenziel 9 Mar 2026 14:23 UTC

  14 points

  2

  in reply to: Daniel Murfet’s comment on: Cole Wyeth’s Shortform

  For those of us who couldn’t make it to Sydney last year—what was your previous position on Solomonoff induction and how did it change as a result of talking with Cole in Sydney?

• Alexander Gietelink Oldenziel 6 Mar 2026 17:05 UTC

  4 points

  0

  in reply to: Cleo Nardo’s comment on: straw­berry calm’s Shortform

  By the way you might also be interested in Aristotle & Boethius’s original work on the topic.

• Alexander Gietelink Oldenziel 6 Mar 2026 17:04 UTC

  5 points

  0

  in reply to: Cleo Nardo’s comment on: straw­berry calm’s Shortform

  What you are talking about are ′ generic objects’ .

  They (re)appear in language/​linguistics, logic and computer science. E.g. in computer science they are closely related to ‘parametric polymorphisms’/​ ′ generic data types’, in logic with Girard’s System F.

  There is a large literature on them. Eg semantics of generic objects is occasionally used to semanticize anaphora, see e.g. here

  The most interesting bit to me is that they are related to a dual connective to universal/​existential quantifiers that are defined by elimination.

  Concretely—if I have to prove \exists \phi(x), the logical introduction rule says that I have to supply an instance t such that \phi(t). I can then conclude \exist \phi(t) by the \exists-introduction rule. In a standard natural deduction style type theory by logical harmony there is a corresponding elimination rule that would eliminate these \exists \phi(x).
  
  However, we could also posit a new quantifier that is defined by it’s elimination rule. So in that case \forall x \phi(x) would produce a c such that \phi(c) iff \forall x \phi(x), in other words a generic object c.

  It’s a complete non-starter. To see what kind of object every-dog is, you can examine it’s properties. For any predicate φ, every-dog satisfies φ iff every dog is φ. For example, every-dog is a mammal, weighs less than 800 tonnes, etc. But every-dog lacks the property of being four-legged, and lacks the property of not-four-legged, since dogs vary on this. Every-dog violates excluded middle — it’s properties are gappy.

  exactly. so this generic c would be your ‘every-dog’. Since we’re working in type theory we are not forced to use excluded-middle so these objects can happily exist without contradiction. =)

  Some-dog is the dual: it has every property instantiated by at least one dog. So it’s simultaneously brown-all-over and white-all-over, male and female, three months old and twelve years old. Some-dog violates non-contradiction — it’s properties are glutty.

  No-dog has every property that no dog has. It’s a prime number. It’s the Eiffel Tower.

  No-dog is not so interesting for non-typed settings. However, if we type or use various universes we can do much richer things.

  rk. More generally in full (Martin-Lof) type theory

  We can extend from \exists and \forall to all \Sigma and \Pi -types [and Universe types]. This would give constructions of arbitrary generic objects and universes satisfying any collection of properties...

  
  Despite being a terrible semantics of quantifiers, this corresponds (if you squint) to the isomorphism between a finite-dimensional vector space and its double dual. Think of the domain of objects D as a vector space, and predicates as linear functionals on D — elements of D*. Then quantifiers live in D**, functionals on predicates. There’s a canonical map D → D** given by x ↦ (f ↦ f(x)): each object corresponds to the quantifier “evaluate at x”, i.e. the proper name quantifier. In finite dimensions this map is an isomorphism — every quantifier is a name in disguise.

  Yes you are definitely on to something here. There are some connections with double-dual constructions. I think the definite treatment is still left to be worked out though.

• Alexander Gietelink Oldenziel 5 Mar 2026 14:39 UTC

  3 points

  0

  in reply to: TsviBT’s comment on: TsviBT’s Shortform

  That is indeed what I imagine you are saying. Perhaps I am inferring too much.

• Alexander Gietelink Oldenziel 5 Mar 2026 9:16 UTC

  2 points

  2

  in reply to: TsviBT’s comment on: TsviBT’s Shortform

  None of these strong enough military, strategic or economic incentives. Sorry, you just can’can’t solve collective action problems by wanting it badly enough. That s not how it works.

• Alexander Gietelink Oldenziel 4 Mar 2026 18:36 UTC

  6 points

  2

  in reply to: Matthew Khoriaty’s comment on: Matthew Kho­ri­aty’s Shortform

  I guess im confused about this conceptually. To me AGI/​ASI by definition does continual learning /​ long-long-horizon planning. What would it even mean otherwise?

• Alexander Gietelink Oldenziel 4 Mar 2026 14:18 UTC

  23 points

  19

  in reply to: Matthew Khoriaty’s comment on: Matthew Kho­ri­aty’s Shortform

  This is Not Only Possible, this is Definitely Going To Happen- the only question is When.

  Will most work that people do look hopelessly myopic and streetlighty? Yeah pretty much

An Align­ment Jour­nal: Com­ing Soon

An­nounce­ment: Iliad In­ten­sive + Iliad Fellowship

• Alexander Gietelink Oldenziel 18 Feb 2026 20:01 UTC

  7 points

  0

  in reply to: Cole Wyeth’s comment on: “We are con­fused about agency”

  How do you define an agent?

  Which of the following entities are agents; if they are agents, what kind?:

  1. a thermostat

  2. a rocket actively steered by a Kalman filter

  3. the membrane of an e. Coli

  4. nematode worm

  5. An ant swarm

  6. a dog

  7. a human

  8. a human that is alert and focused

  9. Amazon, the company

  10. A humanoid robot

  11. A robot arm

  12. the United Kingdom

  13. Claude 4.2

  14. Claude 4.2 being called by an ‘agentic workflow’ app.

  15. the process of Evolution by natural selection

  17. all of humanity over the last thousand years.

  18. AIXI

  19. a Solomonoff inductor

  20. A logical inductor

  21. a Szilard engine, i.e. Maxwell demon

  22. etc

• Alexander Gietelink Oldenziel 18 Feb 2026 19:46 UTC

  2 points

  0

  in reply to: Steven Byrnes’s comment on: The brain is a ma­chine that runs an algorithm

  Wow are you saying Universal Computers were only built in the 70s? Half a century after Turing?

• Alexander Gietelink Oldenziel 18 Feb 2026 19:39 UTC

  7 points

  2

  in reply to: jake_mendel’s comment on: ro­hin­mshah’s Shortform

  Rohin Shah’s position seems perfectly clear to me.

  More research good. AI will speed up research along the whole range of 1-4.

  >> If our most scalable current approaches don’t scale all the way through takeoff, then we (humans) will be able to be keeping up and being critically intellectually involved with developing new alignment techniques when our existing ones stop scaling, at least at the level of granularity of (3) so that we can continue to leverage AIs in ways (1-3) above. We need to keep being able to do this indefinitely, or until we come up with an approach to alignment that does scale as far as capabilities will.
  
  This seems a weird way to phrase things. Or seems to be predicated on some assumptions I find surprising. It seems to suggest that not only that alignment is a continual neverending process [likely] but that finding alignment techniques is a continual neverending process ..!

  If we would analogize ‘alignment’ to ‘steering rockets so they hit their target’ this would suggest that not only that rockets need to be continually steered and course-corrected to hit their target—the entire field of rocket steering needs continual updating and course-correcting. There is a perspective from which this is true since as humanity invents faster and better rockets—there is another perspective from which this is sort of misleading: future rockets will be governed by the Laws of Newton and steering comes down to some variant on Kalman filters and the like.