Channel
@stuffyNLP
On this record: Topic · Growth · Engagement · Reactions · Posts · Citations · Telegram's recommendations · Cite this entry
6,656subscribers
+99 since we began measuring on 6 August 2026
Risers and fallers across the register · movement among entries of 3,162–10,000.
| Telegram ID | -1001957868781 |
|---|---|
| Type | Channel |
| Username | @stuffyNLP |
| Created | Between 1 April 2023 and 31 October 2023 — estimated from Telegram’s id allocation, not measured. How this range is calculated. |
| First recorded | 6 August 2026 |
| Last confirmed live | 18 September 2026 |
| Measurements held | 14 |
| Confirmed unchanged | 1 time, most recently 18 September 2026 |
| On Telegram | t.me/stuffyNLP |
Technology — a classification, not a measurement. An on-box language model (Qwen3.6-35B-A3B-FP8, prompt version 1) read this channel’s own recent posts on 12 September 2026 and assigned it the closest of 31 fixed categories, at 100% confidence. This is a model’s judgement about what the channel is likely to be about, not a fact this register measured the way a subscriber count or a view count is measured — it can be revised on a later pass, and it carries no weight anywhere else on this page. How this classification works, and why it has no browse page of its own yet.
| Measured (UTC) | Subscribers | Change |
|---|---|---|
| 18 Sept 2026, 16:14 | 6,656 | +5 |
| 13 Sept 2026, 21:58 | 6,651 | -2 |
| 9 Sept 2026, 18:16 | 6,653 | -2 |
| 4 Sept 2026, 12:56 | 6,655 | +8 |
| 1 Sept 2026, 00:45 | 6,647 | +12 |
| 28 Aug 2026, 18:48 | 6,635 | +3 |
| 25 Aug 2026, 17:32 | 6,632 | +6 |
| 22 Aug 2026, 23:46 | 6,626 | -2 |
| 19 Aug 2026, 13:12 | 6,628 | +41 |
| 16 Aug 2026, 07:04 | 6,587 | +6 |
| 13 Aug 2026, 00:53 | 6,581 | +9 |
| 10 Aug 2026, 02:02 | 6,572 | +7 |
| 7 Aug 2026, 00:10 | 6,565 | +8 |
| 6 Aug 2026, 07:04 | 6,557 | first reading |
9 posts held, back to 13 July 2026 — the reader has not yet reached the start of this channel’s public history, so older posts may sit further back, unread. Read across 10 pages of Telegram’s post history, 20 posts per page.
Nothing published in the last 30 days. ERR and ER are rolling 30-day measures, so there is nothing to compute — we hold 9 posts for this entry, the most recent from 13 August 2026. An engagement rate over an empty window would be a number about nothing.
189 reactions across 9 posts, in 7 distinct kinds. The most used accounts for 35.4% of them.
| Reaction | Count | Share | Share, drawn |
|---|---|---|---|
| ❤ | 67 | 35.4% | |
| 🔥 | 48 | 25.4% | |
| ❤🔥 | 37 | 19.6% | |
| 👍 | 32 | 16.9% | |
| 🤩 | 3 | 1.59% | |
| 👀 | 1 | 0.529% | |
| 🤮 | 1 | 0.529% |
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 9 of the 9 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 189 reactions in total: the kind of figure the paragraph above means by “a reaction total printed elsewhere on the page”.
Measured over the 9 most recent posts we hold, published 13 July 2026 to 13 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.
Context-1 — поисковый агент, который умеет избавляться от лишнего. Часть 1/2 Авторы сегодняшней статьи заявляют следующую проблему. Когда агент ищет что-то в поиске, то нерелевантные страницы, выданные браузером, всё равно остаются в контексте агента. Чтобы справиться с этим, контекст можно обрезать или суммировать. Ещё одно весьма перспективное направление в этой области — self-editing context, при котором модель …
❤5🔥5❤🔥3🤮1
Как агенты оценивают собственный успех Представим ситуацию: инженеру нужно починить баг в сервисе аутентификации с помощью кодового агента. До начала работ агент сообщает, что вероятность успеха — 72%. По ходу работы меняет мнение на 78%, после всех изменений — 92%, а какой-нибудь AI-ревьюер пророчит положительный результат с вероятностью 85%. Однако по итогу патч, сделанный агентом, не работает — то есть прогнозы о…
❤8🔥8❤🔥3
AgentFold — метод управления контекстом для агентов на длинных задачах На длинных сценариях поиска и анализа веб-агенты либо хранят слишком много сырой информации, либо слишком часто её суммаризируют, что забивает контекст. Авторы сегодняшней статьи предлагают метод AgentFold (GitHub), призванный решить эту проблему, сделав память агента динамической и управляемой самим агентом. Сейчас в веб-поиске используются два…
❤🔥10👍9❤3🔥2
Asynchronous Reasoning: Training-Free Interactive Thinking LLMs Сегодня поговорим о статье, в написании которой принимали участие инженеры Яндекса. Публикация посвящена асинхронному ризонингу, а в её основе лежит метод, описанный в работе Hogwild! Inference: Parallel LLM Generation via Concurrent Attention, поэтому сперва — кратко о ней. Это тоже статья от Yandex Research, а также от HSE и IST Austria. Авторы поста…
🔥13❤3👍2
Тренды из мира бенчмарков на ICML 2026 [2/2] SWE-rebench V2: Language-Agnostic SWE Task Collection at Scale Ребята из Nebius расширили свой SWE-rebench на новые языки: сфокусировались на масштабируемости и пригодности для сбора RL-лёрна. Две ключевые части пайплайна: сборка окружения под каждый репозиторий и отбор задач для тестов. Окружение собирают собственным интерактивным агентом. Он читает README или конфиги …
❤🔥10🔥8❤7🤩1
Тренды из мира бенчмарков на ICML 2026 [1/2] Работ о бенчмарках на ICML традиционно много. По сравнению с прошлым годом, в 2026 стало заметно больше бенчей для агентов. А ещё начали чаще встречаться работы из академии. Объяснение простое. Корпорации вкладывают много сил во внутренние бенчи: делают сами, покупают их у data-labeling-компаний (например, Surge, Mercor, Handshake AI, Toloka) — и, как следствие, такие бе…
👍7❤4❤🔥2🔥1
ICML 2026 — личные впечатления Конференция закончилась, но говорить о ней можно ещё долго. Сегодня личными впечатлениями с нашим каналом поделился старший разработчик команды инфраструктуры обучения Alice AI Владислав Тыцкий. Конференция ощущалась очень масштабной: много людей, огромные залы для докладов, плотное расписание и буквально бесконечное количество постеров. Иногда возникало ощущение, что между интересным…
❤13👍6❤🔥5🔥2
Подборка об RL и ризонинге Рассказываем об улучшении RL для сложных задач, оптимизация в RLVR одной строкой кода (!) и обучении компактной модели для дипресёрча. Reuse your FLOPs: Scaling RL on Hard Problems by Conditioning on Very Off-Policy Prefixes При обучении RL на сложных задачах есть две основные проблемы: 1. Большинство роллаутов — wrong, поэтому положительные примеры для основной части задач не появляютс…
❤11❤🔥4👍4🔥4
Ещё больше классных постеров из Сеула — по следам ICML 2026 RE-TRAC: REcursive TRAjectory Compression for Deep Search Agents Сейчас очень много агентов работает в ReAct парадигме (последовательные reasoning + acting). Авторы считают, что такой подход с длинными линейными цепочками плохо подходит для сложных задач, потому что deep search больше похож на дерево гипотез: модель может наметить несколько веток, но потом…
❤13🔥5👍4🤩2👀1
Showing the 9 most recent of 9 posts we hold for @stuffyNLP. 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.
Republished by
Channels on the register that have forwarded this channel's posts into their own feed.
Built only from forwarded posts we have actually read, on both sides. Coverage is early and deliberately incomplete: a missing link means we have not read the post that would prove it, never that the relationship does not exist. Counts are distinct forwarded posts observed, so they only ever go up as we read more.
Named by 3 registered channels — 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.
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.
The reverse of the list above, and a different kind of signal. This does not require this channel to have ever been asked about directly — each row below is a channel we DID ask Telegram about, whose Telegram-generated list happened to include this one. A channel can appear here with an empty list above it, because being named by someone else’s query is independent of having been queried itself.
This channel appears in 13 seed channels' Telegram-generated recommendation lists in total. Each is Telegram’s list for THAT channel, not this one — see how this is measured.
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 18 September 2026 — this entry's latest reading, not the date you are reading this.
“Душный NLP” (@stuffyNLP), 6,656 subscribers as measured 18 September 2026. Telegram Register, tgregister.com/channel/stuffyNLP.
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.