Food for thought: Bayesian Experimental Designer (Was: Food for thought: Schwartzian Transform)

From	Mild Shock <janburse@fastmail.fm>
Newsgroups	comp.lang.prolog
Subject	Food for thought: Bayesian Experimental Designer (Was: Food for thought: Schwartzian Transform)
Date	2026-09-24 14:32 +0200
Message-ID	<119358v$a3jo$1@solani.org> (permalink)
References	<107cdg8$3ok7g$1@solani.org> <10p44s1$16ps$1@solani.org> <1191b63$6fpr$1@solani.org>

Show all headers | View raw

Hi,

Its really amazing how advanced the 80’s were concerning
the early expert system wave. You already found expert systems
or fault diagnosis systems, that sorted their abduction questions

via a clairvoyant value computation. It refers to a theoretical
benchmark in decision analysis, maintenance optimization,
and information economics. I didn’t find some highly visible

Prolog implementation yet, but I might be mistaken. s(CASP)
doesn’t have it yet right? Neither does Web Prolog mention
the concept yet? LLMs use clairvoyant value as well, obviously,

when they stir the dialog. Maybe as an emergent phaenomenon?

                   ┌─────────────────────────────────────────┐
                   │ HOW LLMS VALUE PERFECT INFORMATION      │
                   └─────────────────────────────────────────┘
                                        │
          ┌─────────────────────────────┴──────────────────────────┐
          ▼                                                        ▼
┌─────────────────────────────────┐ 
┌─────────────────────────────────┐
│       EXPLICIT SCAFFOLDING      │         │       EMERGENT PHENOMENON 
      │
├─────────────────────────────────┤ 
├─────────────────────────────────┤
│ The architecture forces a VPI   │         │ Autoregressive next-token 
      │
│ calculation (e.g., using        │         │ prediction mirrors human 
      │
│ information entropy metrics).   │         │ cost-benefit dialogue. 
      │
└─────────────────────────────────┘ 
└─────────────────────────────────┘
One might spin the idea further and just say the valuation
doesn’t predict some physical system behaviour, but rather
reflects user preferences. See also:

Evaluating User-Agent Collaboration
https://arxiv.org/html/2608.27818v1

LLMs can be also bad clairvoyants:

LLMs Fail to Let Go
https://arxiv.org/pdf/2609.25337

Bye

Disclaimer: While clairvoyant valuation has a heavy math
based literature, that can be based on a couple of approaches
from rational reasoning, the LLM literature might be a little

messy and less known, so the above is what an LLM gave me.
I didn’t select and only did cherry picking, to show the drift…

Mild Shock schrieb:
> Hi,
> 
> Creating a compare/3 on cyclic terms is a fascinating
> topic. Obviously the following implementation would work,
> namely transform a cyclic term into a non-cyclic representation,
> 
> and compare its representation:
> 
> compare_rep(C, X, Y) :-
>      rep(X, A),
>      rep(Y, B),
>      compare(C, A, B).
> 
> Provided rep is injective, if the non-cyclic representation can be
> completely ordered, the original cyclic terms will be also
> completely ordered. One might add further requirements to
> 
> rep, namely that it is conservative, ordering acyclic terms in
> the standard order as require by the ISO core standard. This
> was a discussion on SWI discourse a few months ago. But it
> 
> never adressed the issue how to efficiently sort/2 or keysort/2,
> when the involved lists or pair lists contain cyclic terms. Now
> since AI has become so omniscent, it easily handled me a tip,
> 
> both Gemini(*) and Deepseek(**) did that, and pointed me to the
> Schwartzian Transform. Possibly even more ideal when one
> has ultra fast minimization. The idea is very simple, sketch:
> 
> 1. List' = [ (rep(x),x)  | x e List ]
> 2. List'' = sort_on(π1, List')             %% sort on 1st argument
> 3. Result = [ π2(x) | x e List'']          %% project to 2nd argument
> 
> The benefit when measured against predsort/3, that would use
> compare_rep/2, is that predsort might call O(N log(N)) or more
> comparisons. While the above only does a collation key computation
> 
> once per element, making it O(N). AI being quite a buddy here!
> 
> Bye
> 
> See also:
> 
> Schwartzian transform
> https://en.wikipedia.org/wiki/Schwartzian_transform
> 
> (*)
> https://gemini.google.com/
> (**)
> https://www.deepseek.com/
> 
> Mild Shock schrieb:
>> Hi,
>>
>> How would we do a reverse sorted map?
>>
>> I find in Java:
>>
>> TreeMap(Comparator<? super K> comparator)
>> Constructs a new, empty tree map, ordered
>> according to the given comparator.
>> https://docs.oracle.com/javase/8/docs/api/java/util/TreeMap.html
>>
>> Or in Dogelog Player:
>>
>> tree_new(T):
>> tree_new(T, F):
>> The predicate succeeds in R with a new red-black tree.
>> The binary predicate allows specifying a term compare F.
>> https://www.dogelog.ch/typtab/doclet/book/12_lang/05_libraries/03_util/06_tree.html 
>>
>>
>> Here is an example, using the destructive API. But
>> the same constructor works also for the non-destructive API.
>>
>> ?- tree_new(_T), tree_add(_T, 0rInf, foo),
>>     tree_add(_T, 0rNaN, bar), tree_pairs(_T, L).
>> L = [0rNaN-bar, 0rInf-foo].
>>
>> And now using a comparator modifier, aggregate with a comparator,
>> as a closure. Some Joy of Higher Order logic programming:
>>
>> reverse(C, R, X, Y) :- call(C, R, Y, X).
>>
>> ?- tree_new(_T,reverse(compare)), tree_add(_T, 0rInf, foo),
>>     tree_add(_T, 0rNaN, bar), tree_pairs(_T, L).
>> L = [0rInf-foo, 0rNaN-bar].
>>
>> ?- tree_new(_T,reverse(reverse(compare))), tree_add(_T, 0rInf, foo),
>>     tree_add(_T, 0rNaN, bar), tree_pairs(_T, L).
>> L = [0rNaN-bar, 0rInf-foo].
>>
>> Just toying around with my new NaNs.
>>
>> Have Fun!
>>
>> Bye
>>
>> Mild Shock schrieb:
>>> Hi,
>>>
>>> Functional requirement:
>>>
>>> ?- Y = g(_,_), X = f(Y,C,D,Y), term_singletons(X, L),
>>>     L == [C,D].
>>>
>>> ?- Y = g(A,X,B), X = f(Y,C,D), term_singletons(X, L),
>>>     L == [A,B,C,D].
>>>
>>> Non-Functional requirement:
>>>
>>> ?- member(N,[5,10,15]), time(singletons(N)), fail; true.
>>> % Zeit 1 ms, GC 0 ms, Lips 4046000, Uhr 11.08.2025 01:36
>>> % Zeit 3 ms, GC 0 ms, Lips 1352000, Uhr 11.08.2025 01:36
>>> % Zeit 3 ms, GC 0 ms, Lips 1355333, Uhr 11.08.2025 01:36
>>> true.
>>>
>>> Can your Prolog system do that?
>>>
>>> P.S.: Benchmark was:
>>>
>>> singletons(N) :-
>>>     hydra2(N,Y),
>>>     between(1,1000,_), term_singletons(Y,_), fail; true.
>>>
>>> hydra2(0, _) :- !.
>>> hydra2(N, s(X,X)) :-
>>>     M is N-1,
>>>     hydra2(M, X).
>>>
>>> Bye
>>
>

Thread

VIP0909: VibeCore Improvement Proposal Mild Shock <janburse@fastmail.fm> - 2025-08-11 11:35 +0200
  VIP0111: Does a Map have a Constructor? (Was: VIP0909: VibeCore Improvement Proposal) Mild Shock <janburse@fastmail.fm> - 2026-03-14 18:11 +0100
    100% Prolog Hash map beats SWI C Trie (Was: VIP0111: Does a Map have a Constructor?) Mild Shock <janburse@fastmail.fm> - 2026-03-17 04:07 +0100
    AI Accelerators and ISO Prolog multi-threading (Was: VIP0111: Does a Map have a Constructor?) Mild Shock <janburse@fastmail.fm> - 2026-07-29 15:28 +0200
      Actor/Erlang is dead, no Thread and Mailbox conflation [golang channels] (Was: AI Accelerators and ISO Prolog multi-threading) Mild Shock <janburse@fastmail.fm> - 2026-07-29 15:41 +0200
        Summer Challenge: libSQL = Prolog+Modes [VDBE versus π-WAM] (Was: Actor/Erlang is dead, no Thread and Mailbox conflation [golang channels]) Mild Shock <janburse@fastmail.fm> - 2026-07-30 11:22 +0200
          Work slicing can simulate AbortController (Was: Summer Challenge: libSQL = Prolog+Modes [VDBE versus π-WAM]) Mild Shock <janburse@fastmail.fm> - 2026-08-01 13:23 +0200
      Introducing an asm/1 statement in a Prolog system (Was: AI Accelerators and ISO Prolog multi-threading) Mild Shock <janburse@fastmail.fm> - 2026-08-09 20:21 +0200
      Introducing an asm/1 statement in a Prolog system (Re: AI Accelerators and ISO Prolog multi-threading) Mild Shock <janburse@fastmail.fm> - 2026-08-09 20:25 +0200
      MADV_HAWAII: Aloha from the 300ms Abyss [Exception Handling in Prolog] (Re: AI Accelerators and ISO Prolog multi-threading) Mild Shock <janburse@fastmail.fm> - 2026-09-10 23:42 +0200
        MADV_HAWAII: Aloha from Minimal Terms (Was: MADV_HAWAII: Aloha from the 300ms Abyss) Mild Shock <janburse@fastmail.fm> - 2026-09-14 14:43 +0200
          MADV_HAWAII: Decomposing the Ouroboros Road (Was: MADV_HAWAII: Aloha from the 300ms Abyss) Mild Shock <janburse@fastmail.fm> - 2026-09-20 18:30 +0200
    Food for thought: Schwartzian Transform (Was: VIP0111: Does a Map have a Constructor?) Mild Shock <janburse@fastmail.fm> - 2026-09-23 22:01 +0200
      Food for thought: Bayesian Experimental Designer (Was: Food for thought: Schwartzian Transform) Mild Shock <janburse@fastmail.fm> - 2026-09-24 14:32 +0200
        What will microsoft say, will they buy it? (Was: Food for thought: Bayesian Experimental Designer) Mild Shock <janburse@fastmail.fm> - 2026-09-24 15:34 +0200
          The recursive AI bottom line (Was: What will microsoft say, will they buy it?) Mild Shock <janburse@fastmail.fm> - 2026-09-24 15:38 +0200

csiph-web