дурацкая многозначность многозначная дурашливость

I say, a double metaphor is always a beauty. And if it's realized to be a twin - even more so. -- Man, what the heck are you? -- well, hear this and see for yourself.

Every toddler knows that LLMs have a tunable parameter called "temperature". If you pop it up, the bot starts mumbling. Temperature has a long history in computing. It's a concept borrowed from physics: a particle moves according to a field or a flow that eventually should settle down, but very slowly (who said 'glass'?). So it will usually steadily proceed towards its final position, but a high temperature would allow it occasionally to go against the directive and change the trajectory.

Actually, I do not understand enough physics to explain this model, but some 70 years ago someone brought this seed from physics to algorithm science and it grew into a sizeable garden of search and sampling algorithms, and much more.

How do you find a minimum of a function (e.g., the answer with smallest error)? Think of the graph of your function as a surface. Drop a ball on it somewhere and let it roll, by gravity. It will settle in a point with locally lowest value. But if you allow your ball to drive upward sometimes, you may be able to discover a much deeper valley that lies just across the hill. The frequency of such random movements against gravity is controlled by the temperature. One famous algorithm starts with high temperature to enable hectic initial exploration, and then cools down to converge to a local minimum.

Back to LLMs. Every toddler knows that the LLM acts by computing the next word at each step. But the toddler's younger sibling also knows that it actually computes a distribution, that is, a probability for each possible word. Then it would choose the one with the maximal probability (i.e., minimal error). When the temperature is high, it may occasionally choose a less probable word instead. Higher temperature would cause more frequent appearance of less probable words, and the model will become non-deterministic and rambling. So here is the first prong of the metaphor: high temperature is more jolting motion, fluctuations and randomness. Lower temperature - calm, order and determinism. Coming directly from physics, the best of sciences.

The complementary metaphoric branch is of course human body temperature. Look at J. Doe, a reasonable citizen with sane and predictable behaviours. Let her get some fever - and she starts hallucinating, making unexpected physical and linguistic turns and jerks, becomes irritable and random. Cool a citizen down below 32 deg Celcius - and no nonsense will come of him ever.

So it seems that two distinct senses of temperature both give us a useful analogy. But are they indeed that different? The toddler's nanny is well aware of the fact that our enzymes are tuned to work within a small interval about the normal body temperature. Increase the heat and you get protein conformation changes, unstable or undesired rate of biochemical reactions, shifts in half-life time, etc. In particular, neurons fire in unexpected directions, well-established connections falter and new ones are too easy to create. In the opposite direction, turn it cold, and nothing wants to move anymore, reactions shut down, rigidity crawls in.

Suddenly we see, that these two stories are not unrelated parables coming from different realms of wisdom, but rather a manifestation of a principle of a fundamental nature. We also see how little we understand about the action of such principle on ourselves: i've met no clear and comprehensive account of the adventures of our body and mind under high fever (links, anyone?).
What to do of it? Hell if I know. Once in the past, the person's disposition was explained by the unique blend of his body humours. Maybe since then we've got too much carried away imagining our psyche as a high-level programme, running by an interpreter, inside a sandbox, on a virtual machine, under an operating system, while the physics and chemistry of our body are only the hardware. Maybe the levels of abstraction are not that separated. Just think of it.

the passage below is taken verbatim from a serious contemporary advanced statistics textbook, written by a serious advanced statistics professor, having some serious statistical achievements:
"if you want to predict where in the night sky the planets will be, you can actually do very well with a model where the Earth is at the center of the universe, and the Sun and everything else revolve around it. You can even estimate, from data, how fast Mars (for example) goes around the Earth, or where, in this model, it should be tonight. But, since the Earth is not at the center of the solar system, those parameters don’t actually refer to anything in reality. They are just mathematical fictions. On the other hand, we can also predict where the planets will appear in the sky using models where all the planets orbit the Sun, and the parameters of the orbit of Mars in that model do refer to reality [We can be pretty sure of this, because we use our parameter estimates to send our robots to Mars, and they get there]."

this kinda makes me take back all the praise that i granted to this (seemingly) sublime discipline, or at least put it on ice. but hopefully this one is an innocent lamb, an outlier perhaps?

So i found a spare hour to slowly read the famous "Statistical Modeling: The Two Cultures", - a 23-years-old cornerstone of a certain uncertainty mongering. A highly recommended reading. Looking over it, my immediate association is a russian proverb, which i do not a proper english counterpart of: "Force a blockhead to bow to gods, and he will cruck his skull open."

The main theme is: the classical XX century fox statisticians are addicted to simplistic data models (many disadvantages), and there is a new rising tide of algorithmic models (many advantages), which are structure-agnostic, young, dynamic, practice-oriented, down-to-earth, [add your own, using the ever-blooming imagination of a hair-greying boomer in a desperate attempt to defy aging].

It would be educating to recall how the domain of what today is cursed "data science" looked back then. The machine learning community still struggled to prove their virtue, seeing itself as a kind of half-breed between the classical statistics and computer science/applied maths. It was already clear that they are the champions of accurate predictions, but the actual merit of their successes was yet not fully grasped (is it today, for all the consequences, good and bad?).

Well, fast-forward 20 years, and here we stand - deep models all the way down, and nobody cares about their maths or statistics (i know, there is some serious research there; i've read and listened to some of it - it makes me cry, still waiting for someone to dehydrate my tears).

It is also funny to see how some things change, and some never change. So the author complains that classical data models cannot cope with the challenges of the emerging big data problems, while the old-timers respond by complicating their models, and further notes: "Bayesian methods combined with Markov Chain Monte Carlo are cropping up all over". Wouldn't that give you a cute smile today? And so would the passing remark, that SVMs outperform neural nets (ah...). Yet, he is pretty perceptive about his field: a brand-new Vapnik's theory is mentioned, the typical troubles are referred to (model instability - a.k.a. sensitivity to randomness in data, too many variables to account for/select from, influence and inference lost). Most of those troubles are still haunting the field today, even to a higher degree. And how contemporary is the following citation, from the mouth of the man who has invented Random Forest: "There is an effective method for forming ensembles known as “boosting,” but there isn’t any finite sample size theory that tells us why it works so well"?

It is of course mind-blowing how he overlooks the third way, by which all other natural sciences go: understand the process and encode it in the model; then get helped by statistics. Well, it's a sort of "feature, not bug" - statisticians/data analysts are proud of their agnosticism; if you know your data-generator well, you are not a statistician anymore, you're a mere chemical engineer or whatever. This stupid pride-barrier is what forever slows down the scientific progress most of all, as if these hedges were god-given. Well, surprise, they were erected by your forefathers, and if you're as creative and revolutionary as you self-advertise, maybe give it a peek over. The best of us sometimes do it. (Much of these critics are already present in the first comment attached to this article; it sounds a bit old-fashioned and outdated, but in a good old english style; it is actually impressively fundamental: "Better a rough answer to the right question than an exact answer to the wrong question", "Presumably bootstrap and cross-validation ideas may give here a quite misleading illusion of stability").

To sum up: the prophecy has over-self-fulfilled. The scales tipped over, and today algorithms rule, data models shy. Some would say "this did us much good", others: "much good this did us". I would like to resume on a heavier note as usual, noting an involuntary metaphor by-produced by the author. One of the concepts (multitude of valid models) is nicknamed "Rashomon", after "a wonderful Japanese movie in which four people, from different vantage points, witness ..." etc. The fact that the text is oblivious of the Akutagawa's literary source ('In a Grove') and of the title hijacking made by Kurosawa, provides a sad allegory of the don't-look-back and never-mind-the-details state of the field at present.

За пылью - пыль
За мелью - мель
За солью - соль
За стенкой шкафа спит костыль
За молью моль
У дна нет дна в окне дыра шипит февраль
Мышиный сквот в норе нора
За жалью жаль.
Приватный шорох тихих лап в груди в стене
Не стой на месте не иди не жди ко мне
И чем-то чётным на ступенях начертать
За сонмом - сон,
За пнём - порог
За татем - тать.

 Вот так тянется эта непроварившаяся история.

 "...это был удивительный человек. Однажды он сказал, указав рукой: 'это несущая стена'. И стена, в самом деле, тут же исчезла"

 in my humble judgement, no science is as noble and sublime as Statistics.

 Налетай разбирай креатив покупай: 

1. "AI-Bolit" название для стартапа в медицинском АйТи
2. "Deep Learning" - порнофильм пороговой законности