Telegram RegisterThe public register of Telegram

Channel

Свидетели Градиента

@GradientWitnesses

On this record: Growth · Engagement · What this channel posts · Reactions · Posts · Citations · Cite this entry

668subscribers

+0 since we began measuring on 7 August 2026

Risers and fallers across the register · movement among entries of Under 1,000.

Register entry

Telegram ID-1001951936774
TypeChannel
Username@GradientWitnesses
DescriptionЭтот канал предназначен для общения и обсуждений среди тех, кто как и я, хочет видеть что происходит под капотом у нейросетей, и старается развивать интуицию об их природе и свойствах. Для связи: @kraidiky
CreatedBetween 1 April 2023 and 31 October 2023— estimated from Telegram’s id allocation, not measured. How this range is calculated.
First recorded8 August 2026
Last confirmed live8 August 2026
Measurements held2
On Telegramt.me/GradientWitnesses

Growth

6687 Aug 2026, 12:13 — 668 subscribers8 Aug 2026, 02:34 — 668 subscribers7 Aug 2026, 12:138 Aug 2026, 02:34
2 measurements taken within a single day. Dots are measurements; the straight line between them is drawn to join them, not to claim we know the path taken in between — snapshots are recorded only when a count changes, so gaps mean “no change observed”, never “interpolated”. The vertical axis spans 667–669 and does not start at zero.
Measurement log — every subscribers count we have recorded
Measured (UTC)SubscribersChange
8 Aug 2026, 02:34668no change
7 Aug 2026, 12:13668first reading

Engagement

14 posts held, back to 16 June 2026the reader has not yet reached the start of this channel’s public history, so older posts may sit further back, unread. Read across 1 pageof Telegram’s post history, 20 posts per page.

ERR · 30 days
23.2%
avg views ÷ 668 subscribers
Avg views / post
155
1 post measured
Reaction rate
2.58%
reactions ÷ views · ER floor
Posts in window
1
of 14 held

ERR is average views per post over the last 30 days divided by subscribers, the definition TGStat uses, so this figure is comparable with the one you will see elsewhere. It falls structurally as a channel grows: a high ERR on a small channel and a low one on a large channel describe reach mathematics, not quality. We publish the figure and the sample it came from and pass no verdict on it.

ER is defined industry-wide as (forwards + reactions + comments) ÷ views— note the denominator is views, not subscribers. Telegram’s public web preview carries views and reactions but not forward or comment counts, so the reaction rate above is the reactions term only and is therefore a floor: the true ER for this channel is higher by an amount we have not measured and will not estimate.

What these figures were computed from
WindowRolling 30 days · latest post in window 7 August 2026
Posts held14 (16 June 20267 August 2026)
Views total155
Reactions total4
Forwards / commentsnot exposed by the public surface — not measured, not estimated
Readings taken8 Aug 2026, 02:34 UTC

Views are a single reading per post, taken at the time above. A post published in the last day or two is still accumulating views, which pulls the 30-day average down slightly. That is a property of the standard definition rather than a fault in it, so we keep the definition rather than “correcting” the number into something nobody can reproduce.

Precision. Telegram publishes view counts on its public widget in short form — 8.12K, 3.7M — so any reading at or above 1,000 reaches us rounded to three significant figures, and only counts below 1,000 are exact. Averages and rates derived from them are shown to the same precision rather than to the unit: a figure like 3,701,250 would assert digits nobody measured.

Reaction counts are published per emoji and rounded the same way, so a total below 1,000 is exact and a larger one is a sum that may carry a rounded component from each emoji above 1,000. Because it is a sum, it does not look rounded — read a large reaction total as three significant figures per contributing emoji rather than as the figure it prints.

What this channel posts

Photos
169
Links
73

Lifetime counters from Telegram’s own channel header, read 8 August 2026 — not the date at the top of this page, which is when the subscriber count was last read. Below Telegram’s rounding threshold, so these counts are exact.

Reaction mix

77 reactions across 14 posts, in 8 distinct kinds. The most used accounts for 29.9% of them.

Every reaction kind recorded on the sample, most used first
ReactionCountShareShare, drawn
😁2329.9%
👍1924.7%
🔥1924.7%
56.49%
🤔45.19%
🤯45.19%
🎉22.60%
🙏11.30%

No sentiment is inferred, and none should be read in. This table is ordered by count and by nothing else. Emoji do not carry stable meaning across languages or communities — 🙏 is thanks in one channel and mourning in another — so we publish which ones were pressed and how often, and pass no judgement on what an audience meant by them.

Precision. Telegram publishes reaction counts per emoji and short-forms each one — 4.34K, 1.2M — so any single kind at or above 1,000 reaches us at three significant figures, and only counts below 1,000 are exact. The shares above are ratios of those figures and carry the same error. This is also why the total here can differ slightly from a reaction total printed elsewhere on the page: both are sums of the same rounded parts, taken over samples with different edges.

Coverage. Reactions were read on 14 of the 14 sampled posts in this sample. Summed by Telegram’s own count on each post — not by adding up the per-emoji breakdown above — those same posts carry 77reactions in total: the kind of figure the paragraph above means by “a reaction total printed elsewhere on the page”.

Measured over the 14 most recent posts we hold, published 16 June 2026 to 7 August 2026, using the newest reading held for each. Telegram Stars are excluded: they are a payment, not a reaction, and they have their own section.

Recent posts

7 Aug 2026, 12:16 UTC155 views4 reactionsread 8 August 2026

Я тут копошусь по корпусам тектостов статей и вижу зияющее подтверждение того о чём говрил раньше - придумать механизм и сделать чтобы сообщество его заметило - две очень разные вещи. Помнится на на датафесте год назад я рассказывал про то, что локальных минимумов, не счуществует, а ландшафт на самом деле каньон в котором оптимизатор мечется между стенками. Позже выяснилось, что статья о том, что локальных минимумов

4

13 Jul 2026, 10:57 UTC360 views5 reactionsread 8 August 2026
Photo

Ту самую гроккинговую задачку я уже знаю как облупленную. Подумал я тут на досуге, а что если дать той же сетке задачку посложнее, и дал. Три оппернда в обратной польской записи, все возможные операции, то есть 16 вариантов операций в общей сложности. Модуль взял 47, потому что размеры датасета получились, как куб от модуля - эпохи по 10 секунд. Как обычно 50% в валидейшене. В общем-то я ожидал любого разхвития ситуа

😁5

8 Jul 2026, 15:32 UTC466 views1 reactionsread 8 August 2026
Photo

Что бы такого замутить на юбилей? И на сколько это что-то должно быть заумное. Пишите в комментариях.

1

8 Jul 2026, 09:34 UTC444 views4 reactionsread 8 August 2026
Photo

А вы раньше знали, что для достижения лучшего validation loss ваше расписание lr иногда должно его не уменьшать, а увеличивать?

🤯4

4 Jul 2026, 06:44 UTC656 views9 reactionsread 8 August 2026
Photo

Если вы занимаетесь агентской автоматизацией, и ещё не знаете с какой вероятностью LLM полностью крэшит вашу работу, значит вы ещё не до конца разобрались в агентской автоматизации...

😁5🔥4

1 Jul 2026, 21:04 UTC549 views4 reactionsread 8 August 2026

И вот уже бенчмарк Fable снова наливается соком и актуальностью. :))))

🎉2👍2

28 Jun 2026, 07:12 UTC661 views5 reactionsread 8 August 2026
Photo

Гроккинг другим способом, без weight_decay, кстати, ранг матриц после гроккинга не меняет, уже не использующиеся веса и связи так и продолжают лежать в матрице балластом.

🔥5

27 Jun 2026, 17:43 UTC517 views8 reactionsread 8 August 2026
Photo

Вот это гроккинг на 750-ой эпохе, а вот это - как меняются ранги матриц после гроккинга. После гроккинга значительная часть матриц становится попросту не нужна, и спокойненько деградирует в 2-5 раз. Так что киньте тапочком в того, кто скажет, что современные модели оптимальны, и реально нуждаются в том количестве параметров, которые им необходимы до того, как они правильно решат задачу.

👍5🔥3

27 Jun 2026, 13:31 UTC576 views3 reactionsread 8 August 2026

Сегодня обсуждали интересную возникшую петлю отрицательной обратной связи, для большинства метрикой успеха ИИ является принятое количество мердж реквестов, и исправлений, но их проверкой занимаются те же люди которые делают внедрение. Получается, чем менее качественно они свою работу по проверке того, что им натворил ИИ, тем выше оценка его полезности. В голову приходят радикальные меры, типа мы не принимаем ваш рез

👍1🤔1😁1

19 Jun 2026, 14:34 UTC524 views8 reactionsread 8 August 2026

Только что Клод собрал мне валидную научную статью, соответствующую формальным ограничениям, и общепринятому здравому смыслу с одного четёрехстрочного промпта. Конечно при этом ему было доступна более крупная версия моей статьи, включающая ту же тему, доступно было две папки результатов тестовых прогонов, под двести штук, папка с текстами 67 статей по связанным темам, и беглый анализ совпадающих и не совпадающих те

😁7🤔1

18 Jun 2026, 18:25 UTC515 views6 reactionsread 8 August 2026

Из 67 статей по теме гроккинга 21 прошла на A*, хлебная тема, короче, для тех, чей успех и доход зависят от публикаций.

😁5👍1

18 Jun 2026, 08:42 UTC≈1,130 views10 reactionsread 8 August 2026
File

Оценка Fale-5 с циферками. Если хотите лайкнуть - лучше сначала репостните. Напомню, пока эта штука была доступна я успел прогнать свой бенчмарк, (на который пожёг свои лимиты), но не успел подвести итоги с циферками в руках, только качественные выводы и вот наконец-то подъехало сравнение количественное и подробный анализ использованных стратегий. Спонсором и таким образом соавтором этого выпуска является Chill guy

👍6🔥4

Showing the 12 most recent of 14 posts we hold for @GradientWitnesses. View and reaction counts are the latest single reading for each post, not a live figure, and a recent post is still accumulating both. A view count marked was rounded by Telegram before we ever saw it — t.me prints views in full below 1,000 and to three significant figures above, so ≈1,200,000 means somewhere between 1,150,000 and 1,249,999. Unmarked counts are exact. Text is reproduced from the public post preview and truncated for length.

Citation-graph rank

Citation-graph rank — 842,902 of 1,151,006entries in the measured graph. A weighted position computed from the forward and mention edges below — republished posts weigh more than named mentions — and recomputed periodically, over the whole graph. Published only as this ordinal position, never as a score: a position is a fact, and a score printed beside one channel’s name would read as a verdict this register does not make. The two counts beneath stay separate for the same reason mentions are never summed with forwards anywhere else on this page — a named-by count costs nothing to manufacture. The top 100 by this measure, or how it is computed.

Mentions

Named by 1 registered channel — every channel on the register whose own posts have named this one, by its current username or any other username it currently holds, merged from two separately captured readings of the same fact so a namer caught by only one of them is not missed and a namer both caught is not counted twice. A username this channel has since dropped is not matched — that handle may belong to someone else now, and crediting today’s namer to yesterday’s owner would misattribute it.

Named by

Channels on the register whose posts name this channel's handle.

Names

Channels on the register whose handles appear in this channel's posts.

A mention is a weaker signal than a forward and is counted separately for that reason — naming a channel is not republishing it, and a handle in a post body is easy to place deliberately. The post counts beside each row below are distinct posts in which the handle appeared, from posts we have read on both sides — the “Named by N registered channels” figure above is a different count, of distinct NAMING CHANNELS rather than posts, and is not the sum of the rows under it.

Cite this entry

A live page changes as we take new readings, so a citation should name the measurement it is based on, not just the URL. The line below cites the subscriber count as measured 8 August 2026 — this entry's latest reading, not the date you are reading this.

“Свидетели Градиента” (@GradientWitnesses), 668 subscribers as measured 8 August 2026. Telegram Register, tgregister.com/channel/GradientWitnesses.

Full measurement history, CC BY 4.0. Every reading this register holds for this entry, not just the latest one, as a dated, downloadable record: CSV · JSON. Free to use with attribution to tgregister.com. Each file carries its own generation timestamp, which is the figure to cite for exactly when the data was retrieved.