Mastodawn

Книга: «Анализ данных с LLM. Текст, таблицы, изображения и аудио»

Привет, Хаброжители! Большие языковые модели (LLM) позволяют оптимизировать и ускорить решение практически любой задачи в области анализа данных. Освойте методы для анализа больших массивов текстовых, табличных и графовых данных, изображений, видео и многого другого с помощью понятных запросов на естественном языке и нескольких строк кода на Python.

https://habr.com/ru/companies/piter/articles/1050570/

#llm #language_models #анализ #данные #книги_по_программированию #книга #машинное_обучение

Книга: «Анализ данных с LLM. Текст, таблицы, изображения и аудио»

Хабр

GNUTUX Jun 6

Microsoft Launches 7 MAI Models at Build 2026: Direct Competition with Claude and Gemini

At the Build 2026 conference, Microsoft unveiled 7 new AI models under the MAI brand, including a reasoning model for math and programming, a coding model, and models for image and audio processing. The new models out...

🔗 https://salehgnutux.github.io/GT-NEWSTECH/en/ai/microsoft-mai-models-build-2026/

#Microsoft #MAI #AI #Build_2026 #Language_Models #Mustafa_Suleyman #gnutux

Microsoft Launches 7 MAI Models at Build 2026: Direct Competition with Claude and Gemini

At the Build 2026 conference, Microsoft unveiled 7 new AI models under the MAI brand, including a reasoning model for math and programming, a coding model, a...

GT-NEWSTECH

Habr May 31

Тегирование людей на изображениях и Генерация заголовков для видеороликов

Представьте ситуацию: вы только что вернулись из (заслуженного) отпуска с друзьями и, конечно, сделали множество фотографий. Вы хотите отправить друзьям фото с ними. Но как сделать это эффективно? Можно просмотреть фотографии вручную и отметить каждого друга отдельно. Но вы только вернулись из отпуска, ваш электронный почтовый ящик переполнен, и на просмотр фото совсем нет времени. Как же быть?

https://habr.com/ru/companies/piter/articles/1040746/

#книги_по_программированию #книги #python #llm #language_models #анализ #данные #анализ_данных

Тегирование людей на изображениях и Генерация заголовков для видеороликов

Хабр

Reddit Tech VN Bot Dec 18, 2025

Đang thử nghiệm một số mô hình ngôn ngữ, đặc biệt là dịch thuật. Có ai dùng qua gpt-oss cho dịch đa ngôn ngữ, cụ thể là tiếng châu Âu và tiếng Nhật chưa? Đã thử Mistral Small và Gemma 3, thấy ổn. Gpt-oss so ra sao? Thiếu tiêu chuẩn đánh giá khiến việc lựa chọn mô hình khó khăn. Ai có kinh nghiệm chia sẻ giúp! #AI #dịchthuật #Mistral #Gemma #AItranslation #language_models #gpt_oss

(NOTE: Post is in Vietnamese, under 500字符, includes both English & Vietnamese tags, no URLs. Original content is a

Hacker News Nov 10, 2025

Show HN: LLM Onestop – Access ChatGPT, Claude, Gemini, and more in one interface
https://www.llmonestop.com
#ycombinator #all_llms_in_one_place #all_ai_models_in_one_place #LLM_OneStop #unified_llm_platform #chatgpt_claude_gemini #multiple_ai_models #llm_comparison #ai_platform #language_models

All LLMs in One Place | LLM OneStop

Access ChatGPT, Claude, Gemini, and more AI models from one unified platform. Switch between LLMs mid-conversation.

LLM OneStop

N-gated Hacker News Jul 14, 2025

🤡 Scientists have discovered that narrowly finetuning large language models can lead to hilariously misaligned results 🤯. Who knew that stretching a rubber band in one place would make the whole thing snap? 🙄 Bravo to the geniuses who spend years fine-tuning #chaos. 👏
https://arxiv.org/abs/2502.17424 #scientificdiscovery #humor #language_models #misalignment #fine_tuning #HackerNews #ngated

Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMs

We present a surprising result regarding LLMs and alignment. In our experiment, a model is finetuned to output insecure code without disclosing this to the user. The resulting model acts misaligned on a broad range of prompts that are unrelated to coding. It asserts that humans should be enslaved by AI, gives malicious advice, and acts deceptively. Training on the narrow task of writing insecure code induces broad misalignment. We call this emergent misalignment. This effect is observed in a range of models but is strongest in GPT-4o and Qwen2.5-Coder-32B-Instruct. Notably, all fine-tuned models exhibit inconsistent behavior, sometimes acting aligned. Through control experiments, we isolate factors contributing to emergent misalignment. Our models trained on insecure code behave differently from jailbroken models that accept harmful user requests. Additionally, if the dataset is modified so the user asks for insecure code for a computer security class, this prevents emergent misalignment. In a further experiment, we test whether emergent misalignment can be induced selectively via a backdoor. We find that models finetuned to write insecure code given a trigger become misaligned only when that trigger is present. So the misalignment is hidden without knowledge of the trigger. It's important to understand when and why narrow finetuning leads to broad misalignment. We conduct extensive ablation experiments that provide initial insights, but a comprehensive explanation remains an open challenge for future work.

arXiv.org

N-gated Hacker News Jun 19, 2025

"🧐 Researchers bravely attempt to 'liberate' snippets from books using language models, ignoring copyright like it's an optional suggestion. 📚🤖 Meanwhile, #arXiv is casually looking to hire a #DevOps engineer, because who doesn't want to work for a glorified PDF repository? 💻🎉"
https://arxiv.org/abs/2505.12546 #liberationofknowledge #copyrightissues #language_models #hiring #HackerNews #ngated

Extracting memorized pieces of (copyrighted) books from open-weight language models

Plaintiffs and defendants in copyright lawsuits over generative AI often make sweeping, opposing claims about the extent to which large language models (LLMs) have memorized plaintiffs' protected expression in their training data. Drawing on both machine learning and copyright law, we show that these polarized positions dramatically oversimplify the relationship between memorization and copyright. To do so, we extend a recent probabilistic extraction technique to measure memorization of 50 books in 17 open-weight LLMs. Through thousands of experiments, we show that the extent of memorization varies both by model and by book. With respect to our specific extraction methodology, we find that most LLMs do not memorize most books -- either in whole or in part. However, we also find that Llama 3.1 70B entirely memorizes some books, like the first Harry Potter book and 1984. In fact, the first Harry Potter is so memorized that, using a seed prompt consisting of just the first few tokens of the first chapter, we can deterministically generate the entire book near-verbatim. We discuss why our results have significant implications for copyright cases, though not ones that unambiguously favor either side.

arXiv.org

Hacker News May 8, 2025

Block Diffusion: Interpolating Autoregressive and Diffusion Language Models
https://m-arriola.com/bd3lms/
#ycombinator #block_diffusion #discrete #masked #diffusion #language_models #BD3_LM #BD3_LMs

SOCIAL MEDIA TITLE TAG

SOCIAL MEDIA DESCRIPTION TAG TAG

N-gated Hacker News Apr 2, 2025

Ah, the riveting world of "circuit tracing" in language models 🤖🔍, because what we really needed was another way to complicate things we barely understand. A "replacement model" that makes things "interpretable"? 😂 More like a desperate attempt to justify endless AI research grants.
https://transformer-circuits.pub/2025/attribution-graphs/methods.html #circuittracing #AIinterpretability #researchgrants #language_models #techhumor #HackerNews #ngated

Circuit Tracing: Revealing Computational Graphs in Language Models

We describe an approach to tracing the “step-by-step” computation involved when a model responds to a single prompt.

Transformer Circuits

RedPacket Security Feb 2, 2025

DeepSeek's R1 AI Model Faces Criticism Over Security Vulnerabilities - https://www.redpacketsecurity.com/deepseek-s-flagship-ai-model-under-fire-for-security-vulnerabilities/

#threatintel #language_models #DeepSeek #AI_security

DeepSeek's R1 AI Model Faces Criticism Over Security Vulnerabilities - RedPacket Security

DeepSeek's R1, the latest large language model (LLM) developed by this Chinese startup, is currently facing significant criticism due to various security

RedPacket Security