Well of course, but given that the previously employed technique was blitting, fliedumping memory areas, you can't be more efficient than that. Using XML is for transparency (readability).

(And, note, in context, I regard the ms office file format as lousy. The OpenOffice/LibreOffice format is good.)

And before the benchmark of the finished LLM, it would be interesting to consider the techniques used by LLM producers during training to optimize the "think" chunk quality. I cannot remember any good articles about it now.

Or the "<|think_xhigh|> | <|think_low|> | <|think_off|>" tags: apart from this template detail, it is not immediately clear if reasoning_effort is deterministic (API) or is prompt engineering.

There is only one, only complex, that can be rephrased at will - and which was the intention of the coiner of the expression.

> I was unaware that he coined that term in French (when it may already have been coined differently in English before) [...] Is that where you got it from?

I know it from Books, the good old traditional way. And not necessarily books in French: it is in the works of Paul Johnson, for example. It was not "independently coined by many" - all authors I have state that Sauvy has coined the term.

> By the way, Sauvy "repudiated" his definition in 1989

Surely at some point he may have felt to retract, given that an expression coined to make a point and a good intellectual proposal - "what will be the future and the world role of those that we can now recognize as the non-aligned" - from analytical effort had become a synthetic vulgar twisted label. But the descent into vulgarity was not Sauvy's intention, and certainly there is no intent of attributing "superiority/inferiority" to the parts (as you will have seen from having read the piece on the Observateur).

So again: when somebody mentions Schrödinger's cat or Hegel's cows, you cannot assume that the context be violence or discrimination against animals "as mass interpretation became at some point". You cannot attribute to the speaker unproven intentions especially if dubious, especially if frequentistic. In general; as written, here it is explicitly forbidden.

For the rest, incidentally, I do agree with a tirade against the dire state of civilization in the decadent "western world", especially about the denunciation of «eunuch-like consumerist subservience» (50 years ago a conscious warning and today the token of the dead), but it was uncalled for from the post of flohofwoe.

> [French language]

I hope you are not falling into other assumptions.

And what is the difference? It seems the same ("A got dangerous, causing B to act" vs "A got dangerous, having B decide to act").

Avoidable also in view of the following: have you noticed in the past years a surge of indoctrinated that, at every misdeed from some part P immediately note how some other part S championed it? Judge: "Have you committed rape?" Defendant: "What about all the other rapists?". It's not even a bad argument, it's gut diversion - it's for the distraction of animals.

It is certainly political: "the countries non aligned in the "cold war", currently hindered, will demand a political weight". Why have you quoted me omitting «that will demand a better position»? The whole idea from Sauvy cannot prescind from the parallel from the "Tiers Etat": third as in a different thing politically, third like that large class of poor that demand to emerge. Technically, defined as "the non aligned, the other thing", practically, "something with an historic role important in the conflict and about to explode".

In context of these branches: the academician who introduced the concept did not imply any "superiority" of other parts or other lowly characterization. Obviously disadvantages are noted as they are objective - just like we do when we deal with the world with duly detached analysis. None of them can be translated to vile gut understanding. If the academician introduces a concept of "someidea" and beasts misuse it, if you want to discuss the world remain on the "someidea", if you instead want to be an ethologist focus on the misuse.

If some proponent notices a relevance of some phenomenon to the "someidea", those who noted a "Ah, like the beasts intended" commit a grave base failure in the mandate from the guidelines here to responding to the most plausible interpretation and assuming good faith.

And we are waiting for architectures that do - because it's duly.

And the reading you adopted is a simplified superstition.

The expression 'Tiers Monde' was coined by academician Alfred Sauvy in an article on l'Observateur (14 August 1952), to express that the countries not directly aligned to the "two blocks" are akin to the social "third state" as a worldwide class that will demand a better position.

> Nous parlons volontiers des deux mondes en présence, de leur guerre possible, de leur coexistence, etc., oubliant trop souvent qu’il en existe un troisième, le plus important, et en somme, le premier dans la chronologie. C’est l’ensemble de ceux que l’on appelle, en style Nations Unies, les pays sous-développés. [...] Ce qui importe à chacun des deux mondes, c’est de conquérir le troisième ou du moins de l’avoir de son côté. Et de là viennent tous les troubles de la coexistence. [...] Car enfin ce Tiers Monde ignoré, exploité, méprisé comme le Tiers Etat, veut, lui aussi, être quelque chose.

http://www.homme-moderne.org/societe/demo/sauvy/3mondes.html

The expression 'Tiers Monde', which works as "the part to remember in the opposition of the two blocks at war", is coined mirroring the 1789 pamphlet of Emmanuel-Joseph Sieyès "Qu'est-ce que le Tiers-État ?": «Que demande-t-il [le Tiers-État] ? À devenir quelque chose». And Sauvy stresses that the same rise is coming in the global context.

I'd start again using quantifiers and a more complexive analysis.

You missed the whole point of Taalas HC1: that it is Compute-in-Memory.

> 2. Merging storage and computation // Modern inference hardware is constrained by an artificial divide: memory on one side, compute on the other, operating at fundamentally different speeds. // This separation arises from a longstanding paradox. DRAM is far denser, and therefore cheaper, than the types of memory compatible with standard chip processes. However, accessing off-chip DRAM is thousands of times slower than on-chip memory. Conversely, compute chips cannot be built using DRAM processes. // This divide underpins much of the complexity in modern inference hardware, creating the need for advanced packaging, HBM stacks, massive I/O bandwidth, soaring per-chip power consumption, and liquid cooling. // Taalas eliminates this boundary. By unifying storage and compute on a single chip, at DRAM-level density, our architecture far surpasses what was previously possible.

https://taalas.com/the-path-to-ubiquitous-ai/

But ROM has a massive disadvantage being static. So, either it is cheap and practical "like a CD", or decision making will be forced to do its evaluations.

We have a von-Neumann architecture RAM<->CPU, which is really suboptimal for running current relevant Neural Networks ("RAM<----...---->CPU"). Advantage: flexible.

We have a CiM with Taalas HC1 which has the massive and enabling advantages of running NNs very fast and very energy efficiently.

What could high-speed ROM bring? It must be a good combination of "fast" and "cheap" to to be "interesting" for the market, between those two contenders.

I believe that "practical" as in "replaceable" is also a fundamental property of what we desire in this field: the Processing units are not all there is, also the side-RAM (for context, kv-cache etc.) is a necessary part of the system, so the NN-container is just a piece (which needs expensive co-parts). Whether the NN-container is CiM or not, it will be critical if it can be replaced (like a cartridge, disk, etc.) so that the other parts will not need replacement with it.

And where are the warranties that the solution built be reliable (and optimal, etc.)?.

Which AI? LLMs are coding facilitators and code producers.

A problem is solved when the solution is reliable. Non-deterministic Neural Networks are not reliable. In fact,

> more mechanical[] task

that suggests an expectation of process and procedure, which is still not a capability of current architectures.

Sure, you can ask a brains-deficient operator to perform a huge task, but then you'll have to check the whole product, and that remains not cheap.

llama.app is just an URL (for the "advertisement" webpages of llama.cpp outside GitHub).

> that much better than Ollama

llama.cpp is the real thing, ollama was a fork that remained inferior.