Alexander Gietelink Oldenziel

• Alexander Gietelink Oldenziel 9 Jun 2026 11:58 UTC

  22 points

  21

  in reply to: Austin Chen’s comment on: Daniel Birn­baum’s Shortform

  I agree. This is a major open question.

  From experience, by default people say they donate but then mostly don’t. Even if they donate the money to some foundation it often doesn’t get deployed [ eg Future of Life institute has half a billion in assets since an Ethereum donation and has deployed almost nothing?].

  Dustin & Cari, Jaan are the exception.

• Alexander Gietelink Oldenziel 4 Jun 2026 15:19 UTC

  4 points

  2

  on: An­nounc­ing: Iliad’s Fall 2026 Programs

  What does AoE and EoD mean?

• Alexander Gietelink Oldenziel 4 Jun 2026 9:12 UTC

  6 points

  3

  in reply to: IanWS’s comment on: Agent Foun­da­tions Re­minds Me of Con­ti­nen­tal Philosophy

  >> “Now, setting up sufficiently strong regulation is of course very hard, and one might critique that “proving” that the next generation of a model is aligned is akin to solving alignment itself! But I suspect that guaranteeing a single model is aligned is much easier than solving alignment for all possible models.”
  
  Do you know how to prove ” alignment of a model”?

  This is currently almost as hard as solving alignment for all possible models.

  In fact, strictly speaking it is impossible—the model may have simply encrypted its thoughts and there is no method known to man that can break oneway functions.

  Now maybe we make some assumption that we are in a good world where the AI didn’t cryptographically hide its thoughts. It doesn’t really help. The fundamental problem is that there is no behaviourial tests that can definitely exclude a sharp left turn/​ treacherous turn/​ deception/​ name-du-jour. So there is no ′ proving’, just hope & cope and we’re back to prosaic alignment & evals again.

An­nounc­ing: Iliad’s Fall 2026 Programs

• Alexander Gietelink Oldenziel 25 May 2026 15:36 UTC

  4 points

  0

  in reply to: Leon Lang’s comment on: Your Left Brain Doesn’t Trade With Your Right

  I want to distinguish between

  The context needed

  We want you to do Job A vs we want you to do Job B

  Versus

  You are a worker with experience X [that is crucial for task A] vs you are a worker with Y experience [that is crucial for task B]

  I think no machine, biological or artificial, can consistently perform well on the first question without the context of which job it is being asked to do. This is always the case. [Provably so, I think you can formalize this in mdps]

  However the question is whether it is possible to have a machine that contains both conext X and context Y and dependong on whether it nreds to do task A or task B it can dynamically switch between X and Y. I think this is certainnly possible, frasible, already partially happinging and happining much more in the future for all/​almost all X, rsnging simulatenously over human jobs

• Alexander Gietelink Oldenziel 25 May 2026 11:10 UTC

  2 points

  0

  in reply to: Leon Lang’s comment on: Your Left Brain Doesn’t Trade With Your Right

  Why do you think so ?

  Certainly, it is mathematically the case that there is an exponential number of contexts so platonically your statement is true. But I think the question is whether this is the case for the job contexts that occur in economic life.

  In fact, this is not about job contexts at all but about having any kind of knowledge. You could have the hypothesis that there would be many different AIs for different contexts and skillsets in the past 5 years. But in fact this is not the case. Frontier models are SOTA at (nearly) everything.

• Alexander Gietelink Oldenziel 24 May 2026 8:09 UTC

  2 points

  0

  in reply to: Leon Lang’s comment on: Your Left Brain Doesn’t Trade With Your Right

  Pr3sumably continual learning will solve an issue like catastrophic forgetting.

  You dont have to sharw the entire context- just a summary. How summarizable the context is is something intrinsic about that specific context but in most cases [i.e. not a pseudorandom password phrase] one can lossily compress a lot. Humans hace additional constraints because language is limited, you cajt share genes, you don’t start with some base humans, most parameters cant be altered or shared directly at all. The n3xt phase would be to directly share activation or weight vectors rather than human-interpretable text- that would buy an enormous amount of gain.

  Circling back again to the case of learning different jobs- the “job context” is probably remarkably little, having mostly to do with who the relevant human clients, colleagues, customers etc are and what they want. Note that for vast majority of jobs humans become competent within 0.5-3 years.

  That represents a mix of profession-general skills and context specific for that job. The former the AI will already know- the latter is going to be quite small in most cases.

  Anyway, I feel I’m maybe not getting the point across. I think what I am saying is that arguing for job specific context to matter a lot is a big bet against the bitter lesson. I think that’s unlikely and an epistemic mistake.

• Alexander Gietelink Oldenziel 23 May 2026 23:04 UTC

  2 points

  0

  in reply to: Leon Lang’s comment on: Your Left Brain Doesn’t Trade With Your Right

  I see. So my claim here is that the sifferences between human individuals can be very large because they have a strong limit on context merging and sharing. The full context of an individual is the sum total of their life experience brain development genes etc.

  This limit is far weaker for AI. Merging and sharing context between diffwrent instances of Claude Present or Claude Future is far far easier & orders of magnitude faster and more accurate than between humans.

• Alexander Gietelink Oldenziel 23 May 2026 23:01 UTC

  2 points

  0

  in reply to: Leon Lang’s comment on: Your Left Brain Doesn’t Trade With Your Right

  This seems to be a different objection than that made by chasmani. Not sure what the objection is now.

• Alexander Gietelink Oldenziel 23 May 2026 19:11 UTC

  2 points

  0

  in reply to: chasmani’s comment on: Your Left Brain Doesn’t Trade With Your Right

  Where in the essay did I give that impression?

Your Left Brain Doesn’t Trade With Your Right

• Alexander Gietelink Oldenziel 23 May 2026 13:31 UTC

  29 points

  0

  on: Alexan­der Gietelink Ol­den­z­iel’s Shortform

  Executive Clock Speed

  A concept I’ve found useful: executive clock speed — the rate at which the leadership of an organization can make decent (“80/​20”) decisions that actually get executed.

  Sam Altman used to compare how quickly YC’s best founders answered his emails versus the mediocre ones: “It was a difference of minutes versus days on average response times.” Email latency is a proxy for the founder’s OODA loop, the externally visible signature of internal decisionmaking.

  As organizations scale, this becomes the binding constraint. Most decisions are 80/​20: the cost of waiting a week for the optimal call exceeds the cost of being slightly wrong now. A slow executive forces the whole org to idle against their queue.

  Some consequences of this model:

  • Co-CEOs are bad. Committees are terrible

  • Consistency matters more than peak quality

  • Founder responsiveness is a leading indicator of speed of the whole organization

  Competition between different companies is more often decided by one company deciding & executing many more decisions than doing much smarter decisions. This is especially true if it easy to copy (technical expertise) - which is the default in most industries. I think this is often true—but not always.

  How to speed up?

  • Observe: invest in measurement. Reports, dashboards, regular skip-levels.

  • Orient: filter the measurement intelligently. By default, most executives drown in reports, meetings, emails. Prioritization in what to look at is almost as important as what to decide on.

  • Decide: usually the culprit is perfectionism or fear. The ⁸⁰⁄₂₀ call now beats the ⁹⁵⁄₁₀₀ call in 90% of the time.

  • Act: ensure decisions actually propagate. A decision that doesn’t effectively reach the people executing it is identical to no decision. Apparently, Napoleon spend most of the day writing long and detailed instructions to his officers.
    
    

  What links here?

  • Links #2: 2026/​05 Part 2 by papetoast (31 May 2026 13:41 UTC; 8 points)

• Alexander Gietelink Oldenziel 23 May 2026 13:10 UTC

  10 points

  −12

  on: Alexan­der Gietelink Ol­den­z­iel’s Shortform

  For a while, working on Improving Batteries to tackle the Climate Crisis was one of the coolest, technically interesting & (seemingly) moral calling for a young lad with a technical bent. The CO2 is still being pumped into the air by the gigatons but meanwhile the battery revolution has given way to the drone revolution which is quickly obsoleting all non-drone militaries, handing power to China over NATO, and opening the way for unprecented terrorist attacks.

  Unforeseen consequences are the norm. A lesson for EAs in there.

  EDIT: it appears the central claim of this shortform is wrong.

  https://​​claude.ai/​​share/​​b981a632-dea8-4478-ad69-2f44e928ae58

  Thanks to @Thomas Kwa for pointing this out.

• Alexander Gietelink Oldenziel 23 May 2026 11:50 UTC

  13 points

  6

  on: PLA Daily Trans­la­tion: Reflec­tions on War­fare Brought by AGI

  It’s striking to me that every authority quoted is Western. It brings to mind that “Alexis de Tocquevill, who I’m sure every American has read”—meme

• Alexander Gietelink Oldenziel 22 May 2026 12:04 UTC

  15 points

  12

  on: the­ory up­lift differ­en­tially benefits safety & is underleveraged

  Looking for an underleveraged funding opportunity with differential safety benefits? Deploy funds here.

• Alexander Gietelink Oldenziel 21 May 2026 13:28 UTC

  41 points

  −3

  on: Women should be able to open things

  My pet theory is that jars being too tight is a plot by a shadowly cabal of reactionary foodstuff manufacturers hellbent on recreating traditional gender roles. The freemasons, Don Draper, cadet branches of the house of Valois, and the Michelin man all implicated.

• Alexander Gietelink Oldenziel 13 May 2026 18:46 UTC

  8 points

  0

  on: Op­ti­mi­sa­tion: Selec­tive ver­sus Predictive

  This made me think of Demski’s post on selection vs control: https://​​www.lesswrong.com/​​posts/​​ZDZmopKquzHYPRNxq/​​selection-vs-control

  I’m curious—Would you agree or disagree with the description that prediction is selection internally? i.e. we consider different hypotheses then select the ones that fit best?

The Iliad In­ten­sive Course Materials

• Alexander Gietelink Oldenziel 6 May 2026 16:52 UTC

  27 points

  0

  on: Alexan­der Gietelink Ol­den­z­iel’s Shortform

  AI drones now a reality and a gamechanger on the Ukraine battlefield.

  Apparently, the first AI drones were likely being trialled in late 2025 but it likely took until a few months ago to scale up production and iron out problems. This was likely the reason that Ukraine has been able to regain territoriy for the first time in years [plus slthe shutdown of Russian access to starlink]

  The killzone has now increased from 5 km to ~20-25km and with the new Hornet AI drones could extend up to 150km from the frontline.

• Alexander Gietelink Oldenziel 2 May 2026 9:25 UTC

  −4 points

  0

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

  “you overestimate yourself, sir”