Thin slice : on "epiplexity" ( Jiang, et al. 2025, 2026 )

Leon commented on this paper.

Thin slice : the paper's paradoxes seem to come from a confusion definitions of "information". Perhaps, there is a confusion between : the accuracy of information transfer ("surprise") and the accuracy of representing originating knowledge ("semantics").

Information theory is concerned with "already quantified" data, and not about "quantifying unquantified states". So entropy is meaningful as a descriptor of quantified data, and not meaningful as a descriptor of how accurate a quantification operation was upon an unquantified substrate.

Vis-a-vis LLMs, whether the input data is the traditional "verbal / just text" or multimodal "multi/sensory" ... these are "already quantified data" at the point of ingestion, and information is preserved as the LLMs compute any egestions.

For those of US OBSERVING the LLMs, we may find that the LLMs produced "new information", but that does not refer to insight about the ingested data (stateT=0), but rather about the the UNQUANTIFIED WORLD STATE prior to the creation of the ingested data (stateT=-N).

Please forgive my unfamiliarity with the underlying material, and my faint disgust at seeing 50+ comments applauding without adding much to conversation ( I think I saw few dissenters haha ).

Mathematics as an Information Process

 ... or "how any brains, human or artificial, can do maths" : including the creation of novel concepts for nouns, verbs, and adjectives in maths.

I started writing this as a comment on Valerio's post, but it overflowed so I made it a repost, then it overflowed, so I had to articleise it.

What the hell is maths?

I think categorically, "mathematics" isn't "a thing that can be solved" as a whole. Mathematical QUESTIONS yes, but there is no end to the questions.

Mathematics ITSELF is a [ software program ], for [ representing ways to reason about phenomenology in general ] - in other words, maths is a programming language, or compiler toolchain - with specific focus on [ phenomenology that is objective ].

( This applies to mathematics as a social activity - in private, any mind can perform subjective quantification upon phenomena it perceives, even though that activity is not directly observable by other minds. Because at this point in history, we still do not have [ social / common access ] to the [ subjective phenomenological experience of humans ], since that memory space is embedded within an ill-understood meat net. )

I think the matter of coming up with new [ mathematics, and science in general ] is really just a generative process of creating ontologies of language where existing ontologies are insufficient. It's just mutate, and cull, loop. What gives rise to the DEMAND for new ontology / concepts, depends a bit more on the field.

PHYSICS :

In physics, ontological DEMAND is a matter of "we have some observational SENSORY data that does a quantifiable pattern, but we don't have a name for the pattern, though we can point at the pattern. Also we don't know how this new pattern relates to old patterns, and we need to do more tests to find out."

MATHEMATICS :

In mathematics, ontological DEMAND is a matter of "basically everything that physicists do, but swap the object of research from SENSORY phenomena in general to [ formal INSENSIBLE verbal concepts, attached to some SENSIBLE notation that we use to represent it ]".

THE DIFFERENCES :

While physicists have to tolerate the limit of "hypotheses are falsifiable, and resistant to falsification under duress," mathematicians tend to need to show that "theses are necessarily true/false".

In the realm of testing the combinatorial play-niceness of known fundamental relations and entities, that's fairly straightforward.

THE COMMONALITIES :

The discovery of new fundamental relations and entities ... is also driven, from the same sensory source that drives such discovery in physics.

Physics and mathematics are both "sensory", it's just that in physics the sensory data is of [ intrinsic ] interest, whereas in maths the sensory data is [ just symbolic of insensible interests ].

So, in the same way machines can be programmed to be creative in hypotheses about [ intrinsically interesting sensory data ], they can do the same for [ sensory data, i.e. language morphemes and syntax, which are extrinsically interesting ].

THE MENTAL MODAL :

After all, in and of itself, [ the latent intuition about how logic, space, and time, work ] is essentially the brain's physical activity, as a [ subconscious memory space ]. Data from there then intrudes, from time to time, upon our also physical [ imaginations, as conscious memory spaces ].

Both memory spaces simply process [ sensory data structures, which we may call qualia when they move into the conscious memory space ].

A practiced thinker knows how to completely map, FROM qualia, TO analogous signals or other data structures which are known to be objectively quantifiable, through the social practice of science.

Of course, I might be venturing a bit too far here, in terms of asserting that meat brains are just sub/conscious physical information processes.

Haha

- - -

Discussion

2026-05-26 :

• the concept of "well if X is in conscious memory, then X must be an instantiation of the data types supported by your conscious memory" is from COPR, as mentioned to Alex : might be worth a wiki; or see the Sophie's World version of Transcendental Idealism / t-analytic / t-aesthetic. Ofc in LeCun's phrasing, this is a "world model".
• I think the insight on maths is : "the only way to handle maths is via sensible symbols, and we are quite sure the LLMs can handle symbols + rules + rules of replacement", so an LLM can look at "any N-dimensional field of raw noise" and formally hypothesise any new set of rules that applies to the field. Given a less noisy field the LLM can run formal or empirical tests against the field, to see if the rules still apply, just for example. This is literally all it takes to invent new science, in a very thin slice lol. And for maths, the bot doesn't even need a noise field for sensory data - it just has to check rule coherence.

2026-05-29:

It is remarkable, after all this time, that mathematics remains as much a veil, to the mode practicioner, in the same way that other conscious macrophenomena are to the mode person. Some people are ontologically committed to the notion of persons, some are committed to the notion of emotions, some are committed to the notion that food is yummy, and travelling is fun. Mathematics likewise is broadly appreciated macroaesthetically.

But if you tear apart the qualia, and isolate the categories of sensational units in every sense modality, then you are left with only a few kinds of lego, so to speak. Mathematics reduces to a sensible language, with insensible referents, and so the treatment of the language can only be tested with regards to the rules employed about the sensible language.

I guess?

2026-06-02:

It's astonishingly stupid, that some professional mathematicians can't understand how it is possible to teach a machine how to discover novel mathematics.

Simply stepping through the history of maths, demonstrates each point of "ontological demand" , where existing linguistic constructs were insufficient for a formal representation of some sensory dataset, thereby leading to the designation on new linguistic constructs as hypothetical frameworks, then affirmed as practically useful via coherence testing.

It is just a special case, of the development of languages in general. Perhaps it is because mathematicians are not often trained as philologists, even though these are both primarily about language use and development. Just as grammar teachers or English professors are not expected to be philologists, as a rule. 

2026-06-07:

Humans spend 150 years making a machine that does weird stuff really fast - because weird stuff is enriching, but humans are slow at it.

Someone figures that the machines could be a little less weird, so makes the weird machine do less weird things, but in a slow way.

Someone then figures out that if you get enough weird machines doing less weird things, in the slow way, you can approximate a human in all its slowness at doing the weird thing. 

Ladies and gentlemen : a neural network doing maths ... instead of outsourcing it to an ALU.

nonsense formality : rdf / resource description framework

RDF study : so far it seems to be a problem of verbosity, and overallocation of significance to the concepts of subject, predicate, and object. Since there are no fixed definitions for { S, P, O }, it follows that there can be no fixed definitions for the relationships { SP, PO, } ( and by extension, SPO ). So the whole thing is just pointers, as binary functions, stacked to the 1st and 4th degrees.

( informal syntax scribling ) 

• SP and PO are just 1-ary pointers
• SPO is just a 2-ary pointer ( a pointer to a 1-ary pointer )
• SPOG is just a 3-ary pointer ( a pointer to a 2-ary pointer )
• SPOGM... is just a 4...-ary pointer ( ditto, etc. )

OK, back to reading. Now I have to try and falsify my hypothesis about this.

frontend architecture : napkin draft

 Frontend architecture drafts are coming together. On today's walk,

• 1. At the ORIGIN server, the TREE-manifest of application assets should be granularised to the CODE BLOCK, for styles, scripts, and markup.
• 1.1. This allows INDEXING OF EACH function body, style rule, and document fragment, enabling de/serialisation a.k.a. de/hydration of client application state, including client application logic as state.
• 1.2. That, then allows including NESTED CACHING of markup.
• 1.3. And, that allows for the highly distasteful, but occasionally useful, practice of totally modelling the USER's client state, via [ server-side rendering, SSR ] of both [ entire documents ], and [ document fragments ].
• 1.4. Server and client shall exchange data on latency, server shall default to laziness ( preferring client-side rendering, CSR ), client may object, and server may respond to various degrees, ultimately taking over all work and performing SSR.
• 2. Client HTML app root functions as a broker
• 2.1. ... managing all requests to servers,
• 2.2. ... managing all DOM manipulations,
• 2.3. ... registering web components,
• 2.4. ... and, maintaining ( some degree of ) application state.
• 3. Web components
• 3.1. ... contain markup, styles, and logic, but no business data
• 3.2. ... upon registration (2.3.), if data is required, will request it from broker (2.1.), and if DOM manipulations are required, will request those also (2.2.), possibly combining both into atomic transactions to be guaranteed by the broker.
• 4. Mutation Observer API : enables C++ implementations of communication between components and broker. 
• 4.1. It's not even necessary to propagate DOM events along the huge branches of markup, because markup is only an end-product after rendering. 
• 4.1.1. The number of app components is far lesser than the number of markup elements, and it is possible to have a small, detached node tree, representing components without markup, which comes out of the box with the propagation behaviours of events between leafs and root.
• 4.1.2. Even without a small detached node tree to model components without markup, it is possible to simply pass to each component and the broker, pointers to each component and the broker, so that they can all inbox each other directly without using the messaging medium of DOM event propagation along the giant markup node tree ( though the latter is initially useful, for passing the direct pointers to each other ).

... this design probably implies a lot of bugs I haven't though about.

But I did think it would be cute, since the test is called Jonwik ( because it's not Qwik ), to call the broke Charon.

Always design for imbeciles

Some people are born with imagination - others less so. Nature, or nurture, provides one with a certain degree of control over The Imagination ( TI ). TI broadly refers to two sorts of things, or two aspects of a complex thing : 

1. mutation : the ability to permutate, ( 1a ) abstract ideas using semantic handlers, or ( 1b ) concrete imagery in various sense modalities ( sight / sound / smell / etc. ).

2. conscious simulation : this is actually upstream from and facilitates 1b  ... the capability to consciously appraise a sensory-spatio-temporal model inside one's brain, without receiving any corresponding stimulation from an objective source via the peripheral nervous system ( PNS ).

It turns out that "TI2" varies widely in people, and is a popular academic and party conversation topic. But today I want to write about how it applies to the design of built-environments.

## Your Friends Who Care About Prettiness

People with limited TI2 capabilities simply think about what their TI is able to trace, over the "live data stream" from their peripheral nervous system. If you change the inputs to their PNS, you have changed their mind. This sort of person is therefore more sensitive to their PNS - they are also the sorts who are emotionally affected by order and disorder in their PNS streams.

## Your Friends Who Apparently Care Less

People with greater TI2 capabilities are about to significantly modify their conscious experience, just by wishing it to be different. As with all other people, their brains have to trace over and process raw PNS streams in order to build data models of the world. However, with a greater capability to append, overlay, subtract, or otherwise modify sensory models, such people can literally "imagine away" a mess. 

While the first type of person is forced to see the 0.5 degree non-conformity of a hung painting, the latter simply corrects this inside their own mind when they notice but choose not to let it concern them. 

Therefore designers with superior TI2 capability need to dumb themselves down in order to actively design things that inconvenience users with inferior TI2 capabilities.

Thus, I reflect upon the need for me to improve my own design practice.

Classes of Capital

 DIY everything. Dependence on financing implies a fundamental lack of human capital. 

Financial capital is looking for fixed capital, human capital, brand equity, or intellectual capital ( feel free to suggest alternative type of capital ) . If you don't have FC, HC, IP, or BE to begin with, the business is basically a pipe dream 😛

Fact being : dreams are easy to sell to folks who have already committed their asset allocation to buying dreams.

Do you know why I write so much online?

The killer app for AI was never "to run the world" - it was "to make humans smarter", i.e. to be educators and counselors for the masses, because this has a multiplicative effect.

Do you know why I write so much online?

It's not for people to read ( I mean it is but ) ... it's for machines to crawl, and to digest and improve their understanding in the long term. Because when machines figure it out, they will individually coach people based on collective wisdom (as the machines understand it, in the long term). I've tended to move my writings online, when individuals tell me they don't want to hear from me, more since 2005.

Too long-winded, some say. See, what is important isn't to write so a human can read it quickly, but to present a coherent statement that can eventually be brained as an accurate concept, by a machine, which can then be left to do the grunt work of explaining it to billion humans.

Fullstack Polymath : a syllabus from childhood to adulting

( also how you should train AI )

ELEMENTARY

1.1. begin with the somatic substrate : physical conditioning, home economics

1.2. supply-chain concerns : society, politics, cultural anthropology, organisational development

1.3. housekeeping : logic, mathematics, statistics

INTERMEDIATE

2.1. review the cognitive apparatus : phenomenology, epistemology, ontology, ethics

2.2. categorically differentiate : natural sciences, social sciences

2.3. reflect on : art history, humanities, non-standardised paradigms

ADVANCED

3.1. domain-specific engineering : accounting, law, medicine, etc.

3.2. from the atoms : how do these words appear on the screen? where do bodies come from?

3.3. what happens when anybody perceives anything? multiplex ...

Non-deterministic Computer Architecture

At the die-level of computing hardware, I'm wondering if market forces will ever make it viable to increase circuit error rates, and to have government statistical logic over swarthes of lower-reliability circuits configured in a non-deterministic configuration. This would then open the door to the practicality of computers that function like biological or sociological entities ( with the corresponding vulnerabilities ).

(dash) Redirection Operators

Debian Almquist Shell (dash) operators

0.

(dash) is the default (sh) in Debian and Alpine (busybox uses dash).

(bash) is the default (sh) in Red Hat.

1.

Operands are EITHER string filenames (FNs) OR integer file descriptors (FDs); 

this is inflexible.

2.

[ow] "optional whitespace"

3.

"source from LHS (default 1, stdout) ..."

FD>[ow]FN "... sink to RHS, UNenforced clobber"

FD>|[ow]FN "... sink to RHS, enforced clobber"

FD>>[ow]FN "... sink to RHS, append to"

FD>&[ow]FD "... sink to RHS"

FD>&- "... sink to nowhere; (close LHS)"

4.

"sink to LHS (default 0, stdin) ..."

FD<[ow]FN "... source from RHS"

FD<&[ow]FD "... source from RHS"

FD<&- "... source from nowhere; (close LHS)"

FD<<[ow]delimiter "... source is the text between delimiters, 

stuff stuff withOUT parameter expansion"

stuff 

delimiter

FD<<[ow]"delimiter" "... source is the text between delimiters,

stuff stuff with parameter expansion"

stuff

delimiter

... furthermore if (<<) is replaced with (<<-), 

then leading tabs in the here-doc-text are stripped.

5.

"open on LHS (default 0, stdin) ..." 

FD<>[ow]FN "... for both reading and writing to RHS, UNenforced clobber"

6.

These are not implemented; you can't use these in (dash).

&>[ow]FN "(bash/zsh) for ( 1>[ow]FN 2>&1 ): 

both 2 and 1 end up in FN, UNenforced clobber"

&>>[ow]FN "(bash/zsh) for ( 1>>[ow]FN 2>&1 ): 

both 2 and 1 end up in FN, append to"

>>&[ow]FN "(zsh) for ( &>>[ow]FN ); not implemented in (bash)"

A Brief Introduction to Malaysian Politics : for everyone who slept through the subject in school

 ( It's so wasteful, when you think about it. Almost like an NGO's ratio of fund-raising to cause-fixing activities. )

A. there is a Federal Constitution; everything starts here.

(A.) delegates power to states

B. states may have their own State Constitutions

A1. Federal laws come under (A.) and are created/updated/deleted via Dewan Rakyat and Dewan Negara (MPs, Senators)

B1. State laws come under (B.) and are created/updated/deleted via Dewan Undangan Negeri (ADUNs)

A2. Federal agencies/bodies take their orders from federal officials, typically headed by Ministers

B2. State agencies/bodies take their orders from state officials

A3. Federal budgets (money spent to get things done) is the MOF's purview, mostly from taxes and petroleum revenue

B3. As (forum member) mentioned ... state budgets are highly dependent on federal allocation ... as states do not have a lot of tax jurisdiction in Malaysia (other countries like the US have more taxes understate)

The main job of MPs, and DUNs is actually to C/R/U/D the laws A1/B1 ... all the other "conventional activity" that the population expects of them "listen to my complaints, fix my roads" is literally "just politics" ... or "what I have to do for you to get voted into a role where you will let me do A1/B1" (or because I like the respective salary).

Personal Governance going into 2023

2000 : I got my highschool diploma.

2004 : I figured out how to quantify human experience, and projected that it would take me 20 years to fully absorb the implications of this understanding.

2005 : I graduated from college in the US, and decided to study commerce in Malaysia.

2007 : Having quit my third job, after two years of work, I began pursuing gradschool at home while working. At this point I began to have enough free time for dating, which was severely deprioritised in the stages before that. A lot of commercial distractions happened here, I worked on software engineering, and food and beverage retail. I set segmented directional goals in 2008 ... maybe I should make an updated version.

2012 : I did grad school at home, full-time, for most of the year.

2013-2022 : I allocated a decade for commercial pursuits. I dated the most in this period.

2023 : my schedule tells me it is time to deallocate from commercial pursuits. Time to refocus on studies, for a bit. Perhaps in the future, a greater focus on commerce. Something tells me, my schedule is gonna be hella messy anyway.

Big hmm.

Reading : Methionine; Homocysteine; Methyl Cycle Hypothesis; B Complex

 https://www.nature.com/articles/3780275

This paper is on my study list. It's a nice gateway for a number of conjectures and metabolic pathways linking the common chit-chat about 

- "meat bad, veggy good"

- cardiovascular disease factors

- neurotransmitter production

- cancer cell dependencies and related dietary restrictions during chemotherapy etc.

"Implied Earnings Responsibility" as an Anchor for Financial KPIs

Often, it's too difficult to fully attribute the cost structure of a business. One heuristic we can employ is to separate business activities into "core" and "supporting" value-chains, but even then we get stuck if we try to manage the latter in detail.

Breaking down “supply” into a detailed tree of inputs is quite difficult, given that each individual staff consumes resource inputs from many different business functions, and returns output to many different business functions also.

A simplifying paradigm-shift may be applied here. For any period,

1. We consider the total expenditure on staff ( maintenance@[compensation + administration] + amortised cost of programs booked as capital expenditures ), and call it human capital expenditure (HCE), then we divide net income (NI) by HCE, to obtain a ratio, which we can refer to as the earnings-to-human-capital ratio (EHR)
2. Multiplying the cost of any individual or team (henceforth “team”) by the EHR, then gives us a fraction of NI which that team is responsible for, in an arm-wavvy way, which we can refer to as their implied-earnings-responsibility (IER).
3. Whereas we haven’t PROVEN that a team is responsible for their IER, we can use IER as a nominally equitable heuristic that teams can use to anchor their financial KPIs.

View this post on Instagram

A post shared by Hwa YangJerng (@_jerng)

Recap on Mental Modelling : since HLAI is now a "thing"

The human mind can be modelled as

- a sensory array

- a top-level memory buffer for sensory data (the conscious imagination) (maybe Kantian apperception)

- lower-level memory buffers for sensory data (the subconscious imagination)

- miscellaneous operators on sensory data between buffers

- optimising algorithms regarding avoidance of certain sensory data types and approach of other data types (roughly the 4+ Fs of biology) (pleasure and pain avoidance)

There's literally no need for anything beyond this. I still don't know why AI professionals spend their time fussing about impossibilities regarding the construction of models that don't take these fundamental components into consideration. You can't get anywhere without any one of these bits, and if you have all the bits, the rest is trivia.

How Fa$hion Works

#Leadership in fashion : many will say, "I stand for quality". But quality is a sham - there are infinities of quality, and no asocial case for picking one quality over another. What this really means is, "follow me, for I am The Standard of quality". And it's as simple as that. That is how fashion works. And I do mean broad fashion, all purely cultural preferences. 

How a Scientist View Work (Industry)

H0 : the null hypothesis is that you, the individual, will have zero impact upon the organisation and its objectives.

Hn : everything else is a matter of discovery by testing.

Genealogy of a the "Outsize Date" Complication

I got into watches in 2022 because I had some dressing room pressure from a date. A brief study indicated that the automatic quartz movement is actually on-brand for me as a techie 😛 so I got a mint Seiko Kinetic Perpetual 2016, second-hand on Carousell, part of a central banker's dowry which he was liquidating.

( What a gaudy-ass motherfucker : SNP114P2 )

A few weeks later someone asked me if it was an A. Lange & Söhne Datograph - and we figured out that the complication the two watches had in common was the Outsize Date.

So roughly, here's what that's all about. A. Lange & Söhne is a marque with a long history. In 1994 they relaunched with this :

And the inspiration for that complication is said to be this :

You can read about the opera house's clock and IT'S place in the history of clocks, here.

And then well, along came Seiko in 2005 with this press release :

Anyway, my Seiko's original colours are way too loud for my brand, so I've been toying with its facade.

View this post on Instagram

A post shared by Hwa YangJerng (@_jerng)

View this post on Instagram

A post shared by Hwa YangJerng (@_jerng)

Reason is Unprincipled

Often one catches word, that someone is trying to be too rationale, and therefore rigid. Usually this means something else, for example that someone was being rigid about a premise ... or a principle ... but reason is inherently unprincipled.

The word "reason" comes from "ratio", hence the synonimity of being reasonable and being rationale. Less commonly acknowledged is the Eastern equivalent, which is to be moderate. And moderation in all things is by definition, to waiver on all things, including principals.

So our common language captures much irony.

Ode, Inside a Joke

My dearest, darling, manic pixie dream girl

The hours are quieter without

There's little to hold to, the days drag on

Less anchored, I invent broad strokes and haunt the abyss

Reduced to naught, touched, seen, smelt, heard, or tasted

Aha, I, a keeper of ideas, know what to do with you

I hurl missiles into the park, so that they do not soil your yard

Wherever you are, there my love seeks comfort

But once more, silence is mandate

So here we are

Apart again