7.09.2025 состоялся релизPithon 3.14!
На фоне хайпа про NoGIL всё позабыли про другие фичи. Особенно про Multiple Interpreters, который обещает изоляцию процессов но с эффективностью потоков! На сколько действительно это будет эффективно мы узнаем позже, потому что сейчас это лишь первый релиз с ограничениями и недоработками.
Но что там про NoGIL? Теперь этот режим не экспериментальный, а официально поддерживаемый, но опциональный.
Чтобы запустить без GIL нужна специальная сборка. И перед стартом нужно объявить переменную PYTHON_GIL=0
Для вас я собрал готовый репозиторий где достаточно запустить скрпит, который всё сделает:
▫️ соберет релизный Python 3.14 в новый Docker-образ
▫️ запустит тесты в контейнере (GIL, NoGIL, MultiInterpreter)
▫️ распечатает результаты
Тест очень простой, усложняйте сами)
Вот какие результаты у меня:
=== Running ThreadPoolExecutor GIL ON
TOTAL TIME: 45.48 seconds
=== Running ThreadPoolExecutor GIL OFF
TOTAL TIME: 6.14 seconds
=== Running basic Thread GIL ON
TOTAL TIME: 45.54 seconds
=== Running basic Thread GIL OFF
TOTAL TIME: 4.74 seconds
=== Running with Multi Interpreter
TOTAL TIME: 18.30 seconds
Если сравнивать GIL и NoGIL, то на мои 32 ядра прирост х7-x10 (почему не х32? 🤷). При этом нам обещают что скорости будут расти с новыми релизами.
Режим без GIL похож (визуально) на async, тоже параллельно, тоже не по порядку. Но это не IO! и от того некоторый диссонанс в голове 😵💫, нас учили не так!
Интересно, что чистый Thread работает быстрей чем ThreadPoolExecutor без GIL.
Ну и где-то плачет один адепт мульти-интерпретаторов😭 Теперь нужно искать где они могут пригодиться с такой-то скоростью. Скорее всего своя область применения найдется.
Отдельно я затестил память и вот что вышло на 32 потока:
ThreadPoolExecutor GIL ON
305.228 MB
ThreadPoolExecutor GIL OFF
500.176 MB
basic Thread GIL ON
90.668 MB
basic Thread GIL OFF
472.444 MB
with Multi Interpreter
1267.788 MB
Пока не знаю как к этому относиться)
В целом - радует направление развития!
#release
OpenAI Faces Fifth Copyright Lawsuit Over AI Models
Greetings AI & Law Community! OpenAI finds itself entangled in a fifth copyright lawsuit, with Microsoft also implicated. Filed by US writer Julian Sancton, the lawsuit alleges a blatant disregard for copyrights in training AI models, specifically ChatGPT.
Julian Sancton, a New York Times best-selling author, spent five years and extensive resources researching and writing "Madhouse at the End of the Earth." This lawsuit represents him and other writers who feel their work has been misused.
The crux of the matter is OpenAI and Microsoft using copyrighted works without compensating or seeking permission from authors. The lawsuit contends that these companies not only reproduced millions of works but also built a model that mimics the style and themes of these copyrighted materials.
Unlike previous lawsuits against OpenAI, this one includes Microsoft as a defendant. The claim suggests a close collaboration between the two companies in creating and monetizing AI models like GPT-3 and GPT-4, which underpin various Microsoft services.
The plaintiffs demand compensation for their work and seek a court order prohibiting OpenAI and Microsoft from training their models by infringing on copyrighted material.
This marks the fifth lawsuit against OpenAI by writers, emphasizing a growing legal concern within the AI community.
#AILaw#CopyrightLawsuit#OpenAI#Microsoft#AICommunity
ISO and IEC Make Foundational Standard on Artificial Intelligence Publicly Available
Hello, everyone! The International Organisation for Standardization (ISO) and the International Electrotechnical Commission (IEC) have made ISO/IEC 22989, a critical AI standard, accessible to the public.
This foundational standard defines over 110 essential AI concepts, including 'datasets', 'AI agents', 'transparency', and 'explainability'.
ISO/IEC 22989 provides conceptual guidance for Natural Language Processing (NLP) and Computer Vision (CV) models. Its goal is to establish a shared vocabulary, terminology, and framework for AI concepts, promoting fruitful discussions among stakeholders.
Here's a snapshot of key definitions:
- AI System
- Artificial General Intelligence (AGI)
- AI Auditor
- Robustness
- Explainability
- Predictability
- Transparency
- Trustworthiness
As AI regulations surge worldwide, standardization becomes a vital need, especially in the absence of global alignment in regulatory language.
#AIStandards#AI#ISO#IEC#AICommunity
OpenAI столкнулась с пятым иском в связи с нарушением авторских прав
Приветствуем сообщество ИИ&Право! Компания OpenAI оказалась втянута в пятый судебный процесс в связи с нарушением авторских прав, на этот раз вместе с Microsoft. В иске, поданном американским писателем Джулианом Санктоном, утверждается, что при обучении моделей ИИ, в частности ChatGPT, было допущено вопиющее пренебрежение авторскими правами.
Джулиан Санктон, автор бестселлеров New York Times, потратил пять лет и огромные ресурсы на исследование и написание книги "Дом на краю Земли". Данный судебный иск представляет его интересы и интересы других авторов, которые считают, что их труд был использован не по назначению.
Суть дела заключается в том, что компании OpenAI и Microsoft используют произведения, защищенные авторским правом, не выплачивая авторам компенсации и не запрашивая у них разрешения. В иске утверждается, что эти компании не только использовали миллионы произведений, но и построили модель, имитирующую стиль и тематику этих защищенных авторским правом материалов.
В отличие от предыдущих исков против OpenAI, в этом иске в качестве ответчика фигурирует компания Microsoft. В иске говорится о тесном сотрудничестве двух компаний в создании и монетизации моделей ИИ, таких как GPT-3 и GPT-4, лежащих в основе различных сервисов Microsoft.
Истцы требуют компенсации за свою работу и добиваются судебного решения, запрещающего OpenAI и Microsoft обучать свои модели, нарушая авторские права.
Это уже пятый иск против OpenAI со стороны авторов, что подчеркивает растущую правовую обеспокоенность в сообществе ИИ.
#AILaw#CopyrightLawuit#OpenAI#Microsoft#AICommunity
The Complexity of Regulating Foundation Models in the AI Act
Hello, AI & Law community! Kai Zenner, the Head of Office and Digital Policy Adviser at the Office of MEP Axel Voss, shared his opinion on the OECD website about regulating foundation models in the AI Act.
🔹 The Existing Gap: The proposed AI Act by the European Commission, created before foundation models gained prominence in AI, doesn't explicitly cover these versatile models. Their potential for diverse, unforeseen purposes makes it tricky to fit them into the current product safety approach. The Act's use case approach, limiting AI systems to specific risk classes, is too inflexible for the latest foundation models that can handle various tasks. This creates a regulatory gap that needs to be addressed.
🔹 Positive Progress: The European Parliament has taken a proactive step to tackle this issue by introducing Article 28b, which adds a regulatory layer specifically for foundation models. This article outlines nine essential obligations for developers, including identifying risks, testing, evaluation, and thorough documentation. These measures aim to strike a balance between ensuring safety and fostering innovation in the AI landscape.
🔹 Targeted Approach: A crucial consideration is to avoid putting too much burden on smaller providers while still effectively regulating foundation models. Zenner proposes adopting a systemic approach, targeting only a small number of highly capable and relevant foundation models under the AI Act. This strategy could be similar to how Very Large Online Platforms are designated under the Digital Services Act, ensuring a balanced and efficient regulatory framework.
#AIRegulation#FoundationModels#AIAct#AIInnovation#AICommunity#TechLaw#OECDInsights
Advancing Responsible AI: Insights from the American Chamber of Commerce
Hello, AI community! Today, we bring you valuable insights from the American Chamber of Commerce to the European Union's position paper for the AI Act trilogue negotiations. Let's look into the facts and recommendationstons.
🔹 Defining AI: The Chamber encourages EU decision-makers to align the definition of AI with the OECD's definition. A clear and internationally accepted AI definition fosters multilateral coordination in AI policy.
🔹 Streamlining High-Risk Designation: The paper suggests narrowing the scope of high-risk designation to avoid vagueness. This focused approach ensures that AI technologies with genuine high-risk factors receive appropriate attention while minimizing unnecessary burdens on other AI systems.
🔹 Outcome-Oriented Flexibility: Chapter III requirements largely match current responsible AI practices. However, the Chamber emphasizes the importance of remaining flexible and outcome-oriented.
🔹 Obligations for Foundation Models and General Purpose AI: The paper calls for obligations tailored to foundation models and general-purpose AI.
🔹 Transparency in Artificially Generated Content: Transparency is key when it comes to artificially generated content. The Chamber advocates for measures that promote understanding and disclosure of AI-generated content to safeguard users from potential misinformation and manipulation.
🔹 Harmonized Enforcement and Flexible Standards: To create a robust AI landscape, harmonized enforcement and clear yet flexible standards are necessary.
#ResponsibleAI#AIRegulation#AIAct#AITrilogue#TechPolicy#AICommunity#GlobalAI#AIStandards
Leading the Way in Responsible AI: The EU's Vital Opportunity
Hello, AI enthusiasts! Today, we share a policy brief by Pegah Maham and Sabrina Küspert from the Stiftung Neue Verantwortung. They highlight the EU's significant chance to take the lead in responsible AI development through the AI Act and beyond. But to seize this opportunity, understanding the risks associated with general-purpose AI models is essential.
🔹 Risks from Unreliability: One major concern is the lack of control over AI models' behavior. This unreliability could lead to serious issues like discrimination and stereotype reproduction, spreading misinformation, and violating individual privacy.
🔹 Misuse and Dual-Use AI Models: The dual-use nature of some AI models poses another challenge. While they offer immense benefits, they can also be misused by malicious actors for cybercrime, biosecurity threats, and politically motivated purposes.
🔹 Systemic Risks: As AI rapidly integrates into our society, systemic risks emerge. These risks include economic power concentration and inequality, ideological homogenization, and disruptions from the lagging adaptation of society.
By comprehending and addressing these risks, the EU can establish itself as a global leader in responsible AI development.
#ResponsibleAI#AIRegulation#AIAct#AIChallenges#EUAI#TechPolicy#DigitalFuture#AICommunity
Detailed oil and watercolor hybrid, dark-fantasy meets tropical minimalism, masterpiece quality, ultra-detailed, painterly brushwork, 8K. Side view of a young blonde with very long wavy hair blowing in the wind, graceful and relaxed. Minimal accessories: aviator sunglasses with neon turquoise lenses and a few simple bracelets. Bright tropical palette and bold prints: red bustier and red bi. bottoms. She is standing on a surfboard, poised as if catching a gentle wave. Background: seaside beach with palm trees and rocky outcrops suggesting an exotic resort. Natural materials and earthy tones in props emphasize connection with nature. Light, airy mood conveying the ease of a beach vacation, soft cinematic lighting, subtle watercolor textures blended with detailed oil strokes, shallow depth of field, rich but balanced contrast. Original character, no real people, no logos —ar 3:4 —s 1000 —style raw —v 7
#aiartist#neuralnetworks#deeplearning#aicommunity#aifa
Показать полностью...
*****
Subscribe!⬅️ Click!
Which one is your favorite? 🤔
Please select your preferred character from today's AI-generated anime ensemble!
#AnimeGirl#AiArt#Poll#ChooseYourWaifu#AICommunity#Anime