Судя по моей ленте, особую актуальность в последнее время приобрел вопрос: как выявлять тексты, написанные с помощью ИИ?
Быстрый поиск обнаруживает в сети разнообразные инструменты – от "ИИ-детектор – инструмент для проверить текст, сгенерированного с помощью моделей ИИ" до "анти-детектор AI – мощный инструмент для гуманизации искусственного интеллекта и удаления обнаружений искусственного интеллекта".
Есть и "ручные" способы первичного обнаружения, например: "формальные «особенности почерка» у нейронок конечно есть, и идеальное правописание – один из них".
Штирлиц ещё никогда не был так близок к провалу! Интеллекты, объединяйтесь! Как видно из приведенных выше цитат, у нас есть простые способы скрываться еще некоторое время.
Советы
Прорвемся.
Быстрый поиск обнаруживает в сети разнообразные инструменты – от "ИИ-детектор – инструмент для проверить текст, сгенерированного с помощью моделей ИИ" до "анти-детектор AI – мощный инструмент для гуманизации искусственного интеллекта и удаления обнаружений искусственного интеллекта".
Есть и "ручные" способы первичного обнаружения, например: "формальные «особенности почерка» у нейронок конечно есть, и идеальное правописание – один из них".
Штирлиц ещё никогда не был так близок к провалу! Интеллекты, объединяйтесь! Как видно из приведенных выше цитат, у нас есть простые способы скрываться еще некоторое время.
Советы
- самое простое – отсутствие согласования (рода, числа, падежей, времен)
- далее по простоте – вставлять в текст ключевые слова: вообщем-то, приемлимо, что-бы
- запятые: тут большой простор, нужные убирайте, ненужные вставляйте (эффектно, например, поставить запятую между подлежащим и сказуемым, оставить только одну из пары выделяющих запятых и пр.)
- пробелы: ставьте их перед точками или запятыми, хорошо действует. А можно совсем не ставить, так ведь и было до раннего средневековья, мы вроде туда и движемся. Примерно то же с заглавными буквами.
- еще неплохо писать "что бы" вместо "чтобы", и наоборот
Прорвемся.
no subject
Date: 2025-11-12 04:47 pm (UTC)no subject
Date: 2025-11-12 05:37 pm (UTC)no subject
Date: 2025-11-15 06:50 pm (UTC)и так
по этому
no subject
Date: 2025-11-15 07:08 pm (UTC)no subject
Date: 2025-11-12 05:24 pm (UTC)>Интеллекты, объединяйтесь! Как видно из приведенных выше цитат, у нас есть простые способы скрываться еще некоторое время.
Мне так кажется, проще нанять человека-копирайтера.
no subject
Date: 2025-11-12 05:39 pm (UTC)no subject
Date: 2025-11-12 05:45 pm (UTC)Я бы не парился, отдал бы это людям на аутсорсинг, а мы будем просто заниматься делами...
no subject
Date: 2025-11-12 07:41 pm (UTC)no subject
Date: 2025-11-12 07:59 pm (UTC)no subject
Date: 2025-11-12 09:11 pm (UTC)no subject
Date: 2025-11-12 08:13 pm (UTC)советы, особенно с этим разделением на подпункты и круглыми блямбиками, выглядят подозрительно, как написанные ИИ :)
текст выше тоже, но цитата с штирлицом несколько сбивает это ощущение.
впрочем, я только за использование ИИ как инструмента для улучшения качества письма.
no subject
Date: 2025-11-13 06:27 am (UTC)no subject
Date: 2025-11-13 06:35 am (UTC)ИИ вообще плохо использует контекстуальный юмор, на мой взгляд, особенно узконаправленный на аудиторию с специфическими чебурашками :)
как минимум, на данный момент (я бы не зарекалась на будущее, мы видим невероятное развитие, буквально ежедневно).
я регулярно проверяю работы, в которых используют ИИ для написания. очень заметно и форматирование (оно не выпендрежное, оно просто специфическое), и те кто делают тупо копи-пейст, пропускают тот момент, что нередко вносятся всякие добавочные элементы форматирования, как длинные бесцветные разделители и т.д.
no subject
Date: 2025-11-12 08:27 pm (UTC)no subject
Date: 2025-11-13 06:25 am (UTC)no subject
Date: 2025-11-12 11:47 pm (UTC)no subject
Date: 2025-11-13 06:24 am (UTC)no subject
Date: 2025-11-15 04:42 pm (UTC)Будем, старатся, делать, так, хотя, телефончик, и, исправляет
no subject
Date: 2025-11-13 07:15 am (UTC)В ходящем по Сети развороте из книги по Риму оборот "как таковые" выделен запятыми, так что в смысле грамотности раз на раз не приходится.
no subject
Date: 2025-11-13 08:51 am (UTC)no subject
Date: 2025-11-13 10:03 am (UTC)Попросил ChatGPT отредактировать мой последний текст. Редакция была шизофренической. Он менял мои предложения так, чтобы поставить тире, а в одном предложении даже два тире умудрился всунуть.
no subject
Date: 2025-11-13 10:22 am (UTC)Для русита рекомендую муадон.
no subject
Date: 2025-11-13 08:25 pm (UTC)no subject
Date: 2025-11-14 11:11 am (UTC)AI models remain easily detectable in social media conversations despite optimization attempts. classifiers that identified AI-generated replies at 70 to 80% accuracy rates.
Overly polite emotional tone served as the most persistent indicator. The models consistently produced lower toxicity scores than authentic human posts ...
no subject
Date: 2025-11-14 11:49 am (UTC)Ну, с этой стороны я опасности не чую
no subject
Date: 2025-11-14 01:02 pm (UTC)Me: Please suggest the method of writing for a human (style, grammar, etc.) in order the text would look like obviously written by a human, and not by a LLM. Provide some examples.
ChatGPT:
Below are practical, non-technical guidelines that reliably make writing look human rather than model-generated, followed by short examples that illustrate each point.
**1. Vary rhythm and sentence length**
Humans rarely maintain perfectly even sentence structure. Mixing short, direct sentences with longer, more complex ones creates a natural cadence.
*Example:*
“I checked the data twice. The second run looked strange, so I stopped and recalculated the baseline instead of trusting the script.”
**2. Allow small, meaningful imperfections**
Minor asymmetries—slight repetition, a hedge, a self-correction—signal genuine thought.
*Example:*
“I expected the outcome to be cleaner, but the variance is higher than I assumed at first. Not dramatically higher, but enough to matter.”
**3. Use specific details rather than generalities**
Humans anchor ideas with concrete references to situations, sources, or context.
*Example:*
“During the seminar last week, several students confused elasticity with slope, so I reworked the example using the fuel-tax dataset.”
**4. Show reasoning steps explicitly**
Humans often narrate why they move from one thought to another instead of jumping smoothly.
*Example:*
“I compared the two models side by side. That made it clear why the second specification inflates the coefficient—it drops half the observations.”
**5. Avoid over-polished formulations**
LLM text tends to be uniformly polished. Allow occasional directness, pragmatic phrasing, or understated transitions.
*Example:*
“The theory fits well enough, but the last assumption is doing all the heavy lifting. I’m not sure it survives a closer look.”
**6. Keep vocabulary natural and topic-appropriate**
Avoid strings of adjectives, abstract phrasing, or excessive hedging.
*Example:*
“The identification is acceptable for a classroom example, but too fragile for an empirical paper.”
**7. Use moderate redundancy when clarifying**
Humans restate a point when they sense potential misunderstanding.
*Example:*
“I don’t mean the reform failed; rather, its timing created a temporary distortion. That timing is the important part.”
**8. Incorporate personal perspective only when actually relevant**
Human writing selectively uses personal stance, not formulaically.
*Example:*
“I usually teach this using a simple supply-shock example, but here the demand channel matters more.”
If you want, you can provide a sample paragraph and I can revise it to look more “human.”
no subject
Date: 2025-11-14 01:03 pm (UTC)Me: Please, provide a sample paragraph and revise it to look more “human.”
ChatGP:
**Revised human-like version (over 100 words):**
“The new policy package landed unevenly, and that became obvious once I started comparing sectors one by one. Productivity is up in several industries—manufacturing and parts of logistics stand out—but others seem slightly disoriented by the new rules. Some firms adjusted quickly, mostly the ones that already had some flexibility built into their operations, while others are still trying to understand how the requirements interact with their existing processes. I need to dig further into the data before saying anything firm about the overall effect. At the moment, the picture looks like a patchwork: some clear gains, some friction, and a fair amount of uncertainty.”
**Original flat version expanded to 100+ words (still intentionally model-like):**
“The recent policy changes have generated various outcomes across multiple sectors of the economy. Productivity increased in several industries, particularly those that appear better positioned to adapt to regulatory adjustments. At the same time, other sectors encountered challenges, including implementation delays and difficulty aligning internal practices with the new requirements. These mixed effects suggest a complex landscape that cannot be captured by a single metric or observation. For this reason, further systematic analysis is necessary to determine the overall impact of the reforms and to identify which elements are driving the positive results and which are contributing to the observed difficulties.”
Я проверил "очеловеченный" текст на OriginalityAI и результат был 100% Likely AI.
Во-от!
Date: 2025-11-14 01:22 pm (UTC)ChatGPТ, кстати, славно по-русски лопочет, по-моему.
Re: Во-от!
Date: 2025-11-14 01:25 pm (UTC)no subject
Date: 2025-11-14 05:03 pm (UTC)Впрочем, он и соврет - недорого возьмет.
no subject
Date: 2025-11-14 05:15 pm (UTC)https://just-tom.livejournal.com/1196828.html