Login

Recent searches

No recent searches

Search options

Only available when logged in.
fosstodon.org is one of the many independent Mastodon servers you can use to participate in the fediverse.
Fosstodon is an invite only Mastodon instance that is open to those who are interested in technology; particularly free & open source software. If you wish to join, contact us for an invite.

Administered by:

Server stats:

9.8K
active users

fosstodon.org: AboutStatusProfiles directoryPrivacy policy

Mastodon: AboutGet the appKeyboard shortcutsView source codev4.3.7

#summarization

0 posts0 participants0 posts today
Public
Paul Houle

🤖 Resource-Efficient & Effective Code Summarization

(funny, it's harder to make an AI tell what code does than it is to make one write code...)

arxiv.org/abs/2502.03617

arXiv.orgResource-Efficient & Effective Code SummarizationCode Language Models (CLMs) have demonstrated high effectiveness in automating software engineering tasks such as bug fixing, code generation, and code documentation. This progress has been driven by the scaling of large models, ranging from millions to trillions of parameters (e.g., GPT-4). However, as models grow in scale, sustainability concerns emerge, as they are extremely resource-intensive, highlighting the need for efficient, environmentally conscious solutions. GreenAI techniques, such as QLoRA (Quantized Low-Rank Adaptation), offer a promising path for dealing with large models' sustainability as they enable resource-efficient model fine-tuning. Previous research has shown the effectiveness of QLoRA in code-related tasks, particularly those involving natural language inputs and code as the target output (NL-to-Code), such as code generation. However, no studies have explored its application to tasks that are fundamentally similar to NL-to-Code (natural language to code) but operate in the opposite direction, such as code summarization. This leaves a gap in understanding how well QLoRA can generalize to Code-to-NL tasks, which are equally important for supporting developers in understanding and maintaining code. To address this gap, we investigate the extent to which QLoRA's capabilities in NL-to-Code tasks can be leveraged and transferred to code summarization, one representative Code-to-NL task. Our study evaluates two state-of-the-art CLMs (CodeLlama and DeepSeek-Coder) across two programming languages: Python and Java. Our research tasked models with generating descriptions for Python and Java code methods. The results align with prior findings on QLoRA for source code generation, showing that QLoRA enables efficient fine-tuning of CLMs for code summarization.
#ai#llm#ml
Public
Prompt Engineering Ninja

📚 **AI-Powered Study Guide Summarization!** 🤖✨

Struggling with **long study materials**? Learn how **Prompt Engineering** can help you craft AI queries that generate **concise, structured, and effective summaries**! 🧠🔍

📖 promptengineering.ninja/p/prom

Prompt Engineering Ninja · Prompt Engineering for Summarizing Study GuidesBy Catalin Ciocea
#AI#PromptEngineering#StudySmart
Public
h o ʍ l e t t

→ We’re Doing What Searchbots Can’t
thewalrus.ca/were-doing-what-s

“[Summarization tools incorporated into the search engines are] a boon for people seeking quick answers, but a bane for publishers. Disincentivizing curious users from clicking through to a news site for additional information—a trend called zero-click search—sends less traffic to media outlets that invest in the costly #reporting that #AI machines are scraping, strip-mining, and synthesizing.”

The Walrus · We’re Doing What Searchbots Can’t | The WalrusChatGPT won’t break a story, craft a deft turn of phrase, or tell a compelling narrative
#Summarization#search#news
Public
Habr

Обзор приложения NotebookLM

Приложение под названием NotebookLM ( notebooklm.google.com/ ) было выпущено компанией Google около года назад, и на Хабре было по этому поводу два кратких анонса в прошлом году ( раз , два ). На мой взгляд, оно заслуживает обзора чуть более подробного чем эти краткие сообщения, так что попробую восполнить этот пробел. NotebookLM - это инструмент на основе ИИ, который позволяет относительно быстро, удобно и без лишних телодвижений получить краткий разносторонний обзор (саммари) объемных документов (книг, статей), а также интерактивно взаимодействовать с ними (задавать вопросы, касающиеся их содержания). В моем понимании он представляет собой надстройку над "обычным ИИ-чатом", которому в контекст загружен интересующий пользователя документ. Эта надстройка включает в себя: 1. Набор из нескольких преднастроенных стандартизованных промптов, доступных в один клик и ориентированных на работу с объемными текстами ("Составь мне оглавление", "Составь мне FAQ на основе этого текста", и т.п.) 2. Интерфейсное решение ("карточки-плитки на рабочем столе"), которое по замыслу разработчиков, видимо, должно быть более удобным чем "обычный (линейный) чат" 3. Интерфейс чата, который при взаимодействии с текстом в формате "вопрос-ответ" отображает не только ответы на задаваемые вопросы, но и фрагменты соответствующего исходного текста, а также ссылки на конкретные параграфы полного текста-источника. Посмотрим как это работает

habr.com/ru/articles/839668/

accounts.google.comSign in - Google Accounts
#google#summarization#notebooklm
Public
Charlotte Joanne

📚🤖 Dive into the world of AI-powered summarization! 🧠 The difference between truly summarizing and just shortening text. Understand how LLMs craft concise yet meaningful summaries, and why it's more than just cutting words. ChatGPT and Claude have there say: 🔍💡

"Understanding LLM Summarization
Let’s delve into the difference between summarizing and simply shortening text, as well as addressing those 'gotcha' critiques you mentioned.
Summarizing vs. Shortening
Summarizing involves capturing the main ideas and key points of a text in a concise manner. It's about distilling the essence of the information while ensuring the core message is preserved. This process requires understanding the text deeply and conveying the main themes and conclusions.
Shortening, on the other hand, is merely reducing the length of the text without necessarily maintaining the original meaning or context. This can lead to loss of important details and nuances, making the shortened text less informative or coherent.
Example for Clarification
Imagine a paragraph discussing the importance of holistic health:
Original Text:
Holistic health is an approach to wellness that considers the whole person, emphasizing physical, mental, emotional, and spiritual well-being. Unlike traditional medicine, which often focuses on treating symptoms, holistic health seeks to identify and address the root causes of health issues. By integrating various therapies such as nutrition, exercise, stress management, and alternative medicine, holistic health aims to promote overall balance and harmony in one's life.
Summarized:
Holistic health focuses on overall well-being, addressing physical, mental, emotional, and spiritual aspects. It aims to find and treat the root causes of health issues through integrated therapies like nutrition, exercise, and stress management.
Shortened:
Holistic health considers physical, mental, and spiritual well-being, unlike traditional medicine.
In the summarized version, the main points and intent of the original text are preserved. The shortened version, however, loses significant details and nuances.
Addressing Critiques
When critics argue that summarization is just shortening, they might overlook the nuanced approach of true summarization, which retains the essence and intent of the original content. To ensure summaries are accurate and meaningful, it's essential to:
- Understand the Content Deeply: Read thoroughly and identify the key points.
- Capture the Main Ideas: Focus on what is most important in the text.
- Preserve the Context: Ensure the summary reflects the original intent and message.
- Be Concise: Convey the core message without unnecessary details.
By maintaining this approach, summaries become effective tools for conveying information clearly and succinctly without losing meaning.
Why the Perception Exists
1. Complexity of Language: Human language is intricate, and summarizing involves not just shortening but understanding and conveying the core message, which can be challenging.
2. Output Limitations: Sometimes, the output might appear as just shortened text rather than a true summary, leading to the belief that LLMs don't fully grasp the content.
3. Variability in Quality: The quality of summaries can vary depending on the complexity of the input text and the context provided.
How LLMs Create Summaries
1. Contextual Understanding: LLMs analyze the context and content of the text to identify key points and main ideas.
2. Pattern Recognition: They recognize patterns and themes within the text, which helps in distilling the essence.
3. Conciseness: LLMs aim to convey the main message concisely while preserving essential details and context.
Demonstration with an Example
Let’s take a passage and show how an LLM can summarize it accurately.
Original Text:
Philosophy and psychology are two distinct fields that often intersect. Philosophy is the study of fundamental questions about existence, knowledge, and ethics, while psychology focuses on understanding the human mind and behavior. Despite their differences, both disciplines offer valuable insights into human nature and can complement each other. For example, philosophical discussions about morality can inform psychological studies on ethical behavior, and psychological findings about cognition can enrich philosophical theories of knowledge.
Summarized by an LLM:
Philosophy and psychology intersect in studying human nature. Philosophy addresses fundamental questions about existence and ethics, while psychology examines the mind and behavior. Both fields complement each other, with philosophical discussions on morality informing psychological studies on ethics and psychological findings enriching philosophical theories.
Explanation
- Key Points: The summary captures the main ideas: the intersection of philosophy and psychology, their focus areas, and how they complement each other.
- Context and Essence: The essence of how the two fields interact and enrich each other is preserved.
- Conciseness: The summary is concise, avoiding unnecessary details while maintaining the core message.
Continuous Improvement
LLMs are continually improving in their ability to summarize effectively. Feedback and advanced training help enhance the accuracy and quality of summaries."

I wonder if these models really are to good at sounding sensible now? Has anyone got the time and energy to go through something like this and fact/sanity chest it? I got it from ChatGPT and asked Claue to fact check it - but honestly, who knows?

#AIWriting#MachineLearning#NLP
Public
Habr

Как анализировать тысячи отзывов с ChatGPT? Частые ошибки и пример на реальных данных

В этой статье я расскажу про свой опыт решения рабочей задачи — анализ отзывов о компании от пользователей. Мы разберем возможные ошибки и посмотрим на пример кода и реальных данных. Гайд будет полезен всем, у кого нет большого опыта в анализе данных или работе с LLM через API.

habr.com/ru/articles/821287/

ХабрКак анализировать тысячи отзывов с ChatGPT? Частые ошибки и пример на реальных данныхВ этой статье я расскажу про свой опыт решения рабочей задачи — анализ отзывов пользователей о компании. Мы разберем возможные ошибки и посмотрим на пример кода и реальных данных. Гайд будет полезен...
#llm#gpt#chatgpt
Public
Habr

Автоматизируем поиск ценной информации в групповых чатах Telegram с помощью LLM

Устали мониторить бесконечные групповые чаты в Telegram в поисках важной информации? Решение есть! Пишем компактное приложение на Python, которое будет делать это за нас с использованием LLM.

habr.com/ru/articles/804111/

ХабрАвтоматизируем поиск ценной информации в групповых чатах Telegram с помощью LLMВступление Многие из нас состоят в десятках групповых чатов Telegram, которые заменили нам форумы из нулевых. Это могут быть как чаты по интересам, так и тематические площадки, на которых люди...
#telegram#chatgpt#llm
Public
SIGGEN

As the #GEM team already mentioned, we have endorsed the #data2text and #summarization shared tasks taking place this year: gem-benchmark.com/shared_task

For data-to-text, there are two different datasets and you can choose to work with factual, counterfactual, or fictional versions of the datasets.

For summarization, you can work on Swahili, cross-lingual summarizaion, or summarizing English book chapters.

Interesting challenges with a deadline of 5 April with human evaluations starting on the 6th

gem-benchmark.comGEMBenchmark natural language generation systems with GEM.
Public
Habr

Предсказать ошибку. Как методы оценки неопределенности помогают повышать качество seq2seq-моделей

Всем привет! Меня зовут Артём Важенцев , я аспирант в Сколтехе и младший научный сотрудник AIRI. Наша группа занимается исследованием и разработкой новых методов оценивания неопределенности для языковых моделей. Этим летом мы опубликовали две статьи на ACL 2023 . Про одну из них я уже рассказывал в одном из предыдущих текстов — там мы описали новый гибридный метод оценивания неопределенности для задачи выборочной классификации текстов. Другая же статья про то, как мы адаптировали современные методы оценивания неопределенности на основе скрытого представления модели для задачи генерации текста, а так же показали их высокое качество и скорость работы для задачи обнаружения примеров вне обучающего распределения. Ниже я хотел бы подробнее рассказать об используемых методах и результатах, которые мы получили.

habr.com/ru/companies/airi/art

ХабрПредсказать ошибку. Как методы оценки неопределенности помогают повышать качество seq2seq-моделейВсем привет! Меня зовут Артём Важенцев , я аспирант в Сколтехе и младший научный сотрудник AIRI. Наша группа занимается исследованием и разработкой новых методов оценивания неопределенности для...
#uncertainty_estimation#natural_language_processing#machine_translation
Public
Mott

"Automatic Aspect Extraction from Scientific Texts"

Therefore, the aim of our research is to create a tool for automatic aspect extraction from Russian-language scientific texts of any domain. In this paper, we present a cross-domain dataset of scientific texts in Russian, annotated with such aspects as Task, Contribution, Method, and Conclusion, as well as a baseline algorithm for aspect extraction, based on the multilingual BERT model ...

arxiv.org/pdf/2310.04074.pdf

#LLM#summarization
Public *
Shawn M. Jones, PhD

Yesterday at #TPDL2023 David Pride presented “CORE-GPT: Combining Open Access research and large language models for credible, trustworthy question answering”

Rather than #ZeroShot question/answering, Pride’s team combines the #CORE #OpenAccess dataset with #ElasticSearch to create #FewShot prompts that leverage the strength of combining #search results with the #LLM’s (#GPT) #summarization abilities to produce an answer to a user’s question including citations.

Ref: doi.org/10.1007/978-3-031-4384

David Pride at TPDL2023 is presenting “CORE-GPT: Combining Open Access research and large language models for credible, trustworthy question answering” The current slide is titled “Do LLMs produce accurate citations?”
Continued thread
Public
Shawn M. Jones, PhD

What about that metadata that is present? Grusky et al. (doi.org/10.18653/v1/N18-1065 ) realized that, because page authors create that metadata, it can serve as ground truth to evaluate #Automatic #Summarization.

We analyzed pages from #WebArchiving and saw how this metadata evolved. By 2010 we saw a metadata explosion with the use of #Twitter Cards, Open Graph Protocol, #Facebook Tracking, and more. Things like Twitter cards created a metadata renaissance for HTML.

Ref: doi.org/10.1109/JCDL52503.2021

A streamgraph showing the growth of HTML metadata categories over time. Web pages of the past were filled with mostly crawler instructions and standard HTML metadata fields. By 2010, we see a metadata explosion as Open Graph Protocol, Twitter Cards, Facebook Tracking, and more fill the space.
ExploreLive feeds
About