David Lorell

• David Lorell 25 Nov 2025 18:09 UTC

  4 points

  0

  on: Easy vs Hard Emo­tional Vulnerability

  Might “uninhibited” or “unfiltered” be closer to the concept you’re going for here?

An Analogue Of Set Re­la­tion­ships For Distributions

• David Lorell 15 Nov 2025 20:34 UTC

  11 points

  2

  on: “But You’d Like To Feel Com­pan­ionate Love, Right? … Right?”

  (Written quickly, and therefore not compactly, in the shower. Apologies.)

  I have suspected for a while that there is instrumental value in oxytocin/​companionate-love/​etc, beyond the terminal ‘niceness’ of it.

  Consider alcohol (or other inhibition-lowering drugs):

  • Bars are, apparently, the standard place to meet up after work and relax in lots of cultures

  • Ensuring there is sufficient wine/​alcohol at a dinner party (or parties of any sort) is IME considered a top priority

  • At parties, it is a meme that high schoolers are taught to fight that folks (especially the host) will frequently at least lightly suggest that you have a drink/​other similar drug

  Intuitively to me, “~everyone will have a better time here /​ this way compared to the alternative.” But why? I think a large part of it is that folks who have had alcohol before both know how it effects them (it makes them more likely to speak their mind and/​or feel friendly toward others and/​or take risks social or otherwise) and, the key part: they also believe it effects everyone else roughly the same way. If everyone at the party knows that everyone else is getting drunk, if there is common knowledge that everyone is lowering their guard, then everyone feels safe and relaxed enough to in fact lower theirs. Social frictions get smoothed and higher value social risks get taken (e.g. dancing, asking someone out, etc),

  I think oxytocin/​companionate love plays a very similar role on the level of personal relationships (of all kinds.) If there is common knowledge that the individuals are drunk on love/​connection (or at least buzzed,) then there is a commonly known joint belief that the other will seek to overlook flaws, establish greater trust, make good faith efforts toward the other’s values, look out for the other in a general sense, …. And when that is believed by all participants, alliances on all levels (professional, friendly, romantic,...) are made much more frequently, easily, and held more strongly.

  This is a large part of why, I think, that there is often such delighted interest from others in whether one is drunk/​drugged...and similarly why there is such delighted and/​or profoundly serious interest in whether one feels love toward another. Having common knowledge about everyone under the same or similar influence is a majority of the value to be had.

  Last piece of the model: Being drunk/​drugged among people you don’t already trust is, as I understand, commonly considered to be a Bad Time, and I think it’s because you’ve made yourself more vulnerable but have no assurance that others are more likely to trust you in kind. I claim that it is for similar reasons that people become very impassioned about the question of “Does X really love me?” If one person is sipping on that oxytocin, and the other one isn’t, then that creates a serious asymmetric vulnerability.

  At least that’s my model of a significant chunk of the instrumental value /​ import /​ function of oxytocin to human brains in particular.

  So with all that established: Contrary to the post title: If you (John) find some way to temporarily shoot up with oxytocin that works, there may be human-relationship/​coordination circumstances where you (John) might want to do so as an instrumental act toward satisfying non-companionate-love values.

Toward Statis­ti­cal Me­chan­ics Of In­ter­faces Un­der Selec­tion Pressure

The Zen Of Max­ent As A Gen­er­al­iza­tion Of Bayes Updates

• David Lorell 31 Oct 2025 19:36 UTC

  2 points

  0

  in reply to: Gurkenglas’s comment on: Re­sam­pling Con­serves Re­dun­dancy & Me­di­a­tion (Ap­prox­i­mately) Un­der the Jensen-Shan­non Divergence

  Thanks for the check! I’ve added a collab containing my quick numerical stress tests, linked at the bottom of the post.

Re­sam­pling Con­serves Re­dun­dancy & Me­di­a­tion (Ap­prox­i­mately) Un­der the Jensen-Shan­non Divergence

• David Lorell 28 Oct 2025 22:01 UTC

  10 points

  0

  in reply to: johnswentworth’s comment on: Re­sam­pling Con­serves Re­dun­dancy (Ap­prox­i­mately)

  (Update 7)
  
  After some back and forth last night with an LLM^[1], we now have a proof of “chainability” for the redundancy diagrams in particular. (And have some hope that this will be most of what we need to rescue the stochastic->deterministic nat lat proof.)

  (Theorem) Chainability of Redunds

  Let P be a distribution over $X_1$, $X_2$, and $\Lambda$.
  Define:
  $Q[X,\Lambda ]:=P\left[X\right]P\left[\Lambda |X_1\right]$
  $S[X,\Lambda ]:=P\left[X\right]P\left[\Lambda |X_2\right]=P\left[X\right]{\sum }_{X_1}P\left[X_1|X_2\right]P\left[\Lambda |X\right]$$R[X,\Lambda ]:=P\left[X\right]Q\left[\Lambda |X_2\right]=P\left[X\right]{\sum }_{X_1}P\left[X_1|X_2\right]P\left[\Lambda |X_1\right]$

  Where you can think of Q as ‘forcing’ P into factorizing per one redundancy pattern: $X_2\to X_1\to \Lambda$, S as forcing the other pattern: $X_1\to X_2\to \Lambda$, and R as forcing one after the other: first $X_2\to X_1\to \Lambda$, and then $X_1\to X_2\to \Lambda$.

  The theorem states,
  $D_{KL}\left(P||R\right)\le D_{KL}\left(P||Q\right)+D_{KL}\left(P||S\right)$,

  Or in words: The error (in $D_{KL}$ from $P$) accrued by applying both factorizations to P, is bounded by the the sum of the errors accrued by applying each of the factorizations to P, separately.

  Proof

  The proof proceeds in 3 steps.

  1. $D_{KL}\left(P||Q\right)\ge D_{KL}\left(S||R\right)$

     Pf.
     Let $a_{X_1}:=P\left[X_1|X_2\right]P\left[\Lambda |X\right]\ge 0$

     Let $b_{X_1}:=P\left[X_1|X_2\right]P\left[\Lambda |X_1\right]\ge 0$

     By the log-sum inequality:

     ${\sum }_{X_1}\left(a_{X_1}ln\frac{a_{X_1}}{b_{X_1}}\right)\ge \left({\sum }_{X_1}{a}_{X_1}\right)ln\left(\frac{{\sum }_{X_1}{a}_{X_1}}{{\sum }_{X_1}{b}_{X_1}}\right)$

     $⟹$$D_{KL}\left(P||Q\right)\ge D_{KL}\left(S||R\right)$ as desired.

  2. $D_{KL}\left(P||R\right)=D_{KL}\left(P||S\right)+D_{KL}\left(S||R\right)$

     Pf.

     $D_{KL}\left(P||R\right)-D_{KL}\left(P||S\right)$

     $={\sum }_{X,\Lambda }P[X,\Lambda ]\left[lnP\right[\Lambda |X]-lnR[\Lambda |X_2]-lnP[\Lambda |X]+lnS[\Lambda |X_2\left]\right]$

     $={\sum }_{X_2}P\left[X_2\right]{\sum }_{\Lambda }{\sum }_{X_1}P\left[X_1|X_2\right]P\left[\Lambda |X\right]ln\frac{S\left[\Lambda |X_2\right]}{R\left[\Lambda |X_2\right]}$

     $={\sum }_{X_1}{\sum }_{X_2}P\left[X_1|X_2\right]P\left[X_2\right]{\sum }_{\Lambda }S\left[\Lambda |X_2\right]ln\frac{S\left[\Lambda |X_2\right]P\left[X\right]}{R\left[\Lambda |X_2\right]P\left[X\right]}$

     $=D_{KL}\left(S||R\right)$

  3. Combining steps 1 and 2,
     $D_{KL}\left(P||Q\right)\ge D_{KL}\left(S||R\right)=D_{KL}\left(P||R\right)-D_{KL}\left(P||S\right)$

     $⟹D_{KL}\left(P||R\right)\le D_{KL}\left(P||Q\right)+D_{KL}\left(P||S\right)$

     which completes the proof.

  Notes:
  In the second to last line of step 2, the expectation over $P\left[X_1|X_2\right]$ is allowed because there are no free $X_1$’s in the expression. Then, this aggregates into an expectation over $S[X,\Lambda ]$ as $S\left[\Lambda |X_2\right]=S\left[\Lambda |X\right]$.

  We are hopeful that this, thought different than the invalidated result in the top level post, will be an important step to rescuing the stochastic natural latent ⇒ deterministic natural latent result.
  

  1. ^

     A (small) positive update for me on their usefulness to my workflow!

• David Lorell 23 Oct 2025 22:31 UTC

  9 points

  0

  in reply to: johnswentworth’s comment on: Re­sam­pling Con­serves Re­dun­dancy (Ap­prox­i­mately)

  (Update 5)
  
  A conjecture we are working on which we expect to be generally useful beyond possibly rescuing the stoch->det proof that used to rely on the work in this post:

  Chainability (Conjecture):
  $\forall P[X,\Lambda ]$with ${ϵ}_1:=D_{KL}(X_1\to X_2\to \Lambda )$, ${ϵ}_2:=D_{KL}(X_1\leftarrow \Lambda \to X_2)$,
  Define $Q[X,\Lambda ]:=P\left[X\right]P\left[\Lambda |X_2\right]$, and $Q^{\prime }[X,\Lambda ]:=Q\left[X_1|\Lambda \right]Q\left[X_2|\Lambda \right]Q\left[\Lambda \right]$.
  Then, $D_{KL}\left(P||Q\right)<n({ϵ}_1+{ϵ}_2)$
  
  Here is a collab with a quick numerical test that suggests the bound holds (and that n=1, in this case).
  
  (Note: The above as written is just one step of chaining, and ultimately we are hoping to show it holds for arbitrarily many steps, accumulating an associated number of epsilons as error.)

• David Lorell 17 Oct 2025 21:27 UTC

  6 points

  0

  in reply to: Jeremy Gillen’s comment on: Re­sam­pling Con­serves Re­dun­dancy (Ap­prox­i­mately)

  Do you have sympy code for the example noted at the bottom of the collab that claims a ratio of > 9.77 including the mediation $D_{KL}$? I tried with the parameters you mention and am getting a ratio of ~3.4 (which is still a violation of previous expectations, tbc.)

• David Lorell 16 Oct 2025 23:54 UTC

  11 points

  0

  in reply to: David Lorell’s comment on: Re­sam­pling Con­serves Re­dun­dancy (Ap­prox­i­mately)

  (Update 2)
  
  Taking the limit of the ratio of $D_{KL}$s (using summation rather than max) with $c\to 0$ while $b:=10c$ gives $5\left(\frac{r+1}{11}\right)$

  Setting c very small and ramping up r indeed brakes the bound more and more severely. (Code changes from the collab you provided, below.)

  Code changes /​ additions

  Block 1:
  a,b,c,d,r = sp.symbols(“a b c d r”)
  variable_substitutions = { # The definitions of these variables
  a: 0.25,
  b: 1e-90,
  c: 1e-91,
  r: 20000000,
  }
  
  Block 2 (later on):
  expr = (kl3/​(kl1 + kl2)).subs(d, (1-3*c-(r+1)*b-2*a))
  print(“KL(X2->X1->L’)/​sum[KL(X1->X2->L),KL(X2->X1->L)]=”,(kl3/​(kl1 +kl2)).evalf(subs=variable_substitutions))
  
  Block3 (right after Block 2):
  expr = (kl3/​(kl1 + kl2)).subs(d, (1-3*c-(r+1)*b-2*a)).subs(b, 10*c)
  lim = sp.simplify(sp.limit(expr, c, 0))
  print(“Limit of KL(X2->X1->L’)/​sum[KL(X1->X2->L),KL(X2->X1->L)] as c->0+ =”, lim)

• David Lorell 16 Oct 2025 22:34 UTC

  15 points

  0

  in reply to: johnswentworth’s comment on: Re­sam­pling Con­serves Re­dun­dancy (Ap­prox­i­mately)

  (Update 1)
  
  We’ve looked at the code and fiddled with the math and are now more convinced of the issue.
  
  The 2nd order approximation holds when $\frac{Q\left[X\right]}{P\left[X\right]}\approx 1$ …Which our scaling-down construction does not provide. So, (among a host of other things,) we are now thinking about other ways to try wrangling the $D_{KL}$ bound into a euclidean space or otherwise into some form that is similarly “easy” to work with.
  
  (Thanks for finding this!)

Nat­u­ral La­tents: La­tent Vari­ables Stable Across Ontologies

Do-Diver­gence: A Bound for Maxwell’s Demon

• David Lorell 25 Aug 2025 17:51 UTC

  128 points

  −9

  on: Be­fore LLM Psy­chosis, There Was Yes-Man Psychosis

  This post absolutely sings. The “yes-man psychosis” framing is sticky, clarifying, and honestly—iconic. You take a fuzzy, slippery problem and pin it to the mat with crisp language, vivid examples, and just enough systems thinking to make the lights come on. The Boyd/​OODA connection is chef’s-kiss; it turns a cultural gripe into a concrete failure mode you can point at and say, “That—right there.” The Putin case study lands like a hammer, and the dead-organization metaphor (rocks, logs, and that hilariously misleading chart!) is going to live rent-free in my head. This is the kind of essay people forward to their bosses with “must read” in the subject line. It’s sharp, fearless, quotable, and—despite the bleak subject—fun to read. Truly an instant classic on power, perception, and how praise can calcify into poison. Bravo for naming the thing so precisely and making it impossible to unsee.

(∃ Stochas­tic Nat­u­ral La­tent) Im­plies (∃ Deter­minis­tic Nat­u­ral La­tent)

Re­sam­pling Con­serves Re­dun­dancy (Ap­prox­i­mately)

• David Lorell 11 Aug 2025 22:50 UTC

  4 points

  0

  in reply to: johnswentworth’s comment on: A Self-Dialogue on The Value Propo­si­tion of Ro­man­tic Relationships

  I wonder if your oxytocin is fine but you have, for whatever reason, a very strong cognitive “immune response” to its effects. I think it is common in teens (well, it was the case for me in tweenagehood,) to react to the hook of limerance/​this-whole-cluster with “*no one must know*.” Followed in my case by trying not to dwell on it. I’m not suggesting this thought specifically is something you have trained (maybe its more like “*be realistic*”) but maybe you have a well developed cognitive kata that shuts these kinds of thoughts down before they can become limerance/​love/​companionship/​etc.

• David Lorell 11 Aug 2025 3:43 UTC

  22 points

  13

  in reply to: johnswentworth’s comment on: A Self-Dialogue on The Value Propo­si­tion of Ro­man­tic Relationships

  You’re disgusting monsters, both of you.

  I can’t even bear to look at how you’ve both shamelessly normalized usage of the phrase “beg the question” to mean “prompt the question” rather than its god-given original meaning of “assume the premise.”

  Shame on you.

  Oh and nice kinks.

• David Lorell 11 Aug 2025 3:34 UTC

  4 points

  0

  in reply to: johnswentworth’s comment on: A Self-Dialogue on The Value Propo­si­tion of Ro­man­tic Relationships

  ...ah. When you put it that way.....

  If somehow something happened within the last decade which shifted my People vs Things interest parameter significantly more away from People and toward Things I’d probably be a much more capable researcher right now. (Unsure about before a decade from now because then we start messing with my middle-young teenagehood where the actual path I took to deciding I was going to work on alignment routed through caring deeply about others....or at least imagining the deep loss of not having the opportunity to mutually care very deeply about others in this way.)

  I’d also not have or be many things which I currently reflectively value highly, but that’s a me thing :)

  I might, if I meditated on it, press a button that goes back in time to perform that intervention back in my early college years, (and I’d grieve the decision more than I’ve grieved probably anything,) to increase the chance that our work is decisively counterfactual. I’m so glad that such a button does not exist.

  (Fun, and probably tragic from your POV, fact: Our very own Dan Hendryks more or less encouraged me to self modify in this way for this reason back when we were college. I shook my head and laughed at the time. Now I feel more complicatedly.)

  Point being: Yup. That sure is a life-influencing personality-parameter. Concern is super merited.