Kỹ thuật suy luận thích ứng độ sâu trên mô hình Mixtral: thay vì dùng 32 lớp, hệ thống chỉ kích hoạt ~24 lớp động tùy độ phức tạp của đầu vào. Giảm đáng kể tính toán mà không cần huấn luyện lại, mở ra hướng tối ưu hóa AI mới.

#AI #LLM #Mixtral #Optimization #TốiƯuHóa

https://www.reddit.com/r/LocalLLaMA/comments/1q66ao2/depthadaptive_inference_on_a_mixtral_backbone_32/

Kann man gedruckte Ortsfamilienbücher mit KI auswerten?

Dass Ortsfamilienbücher exzellente Forschungsgrundlagen für die Geschichts-, Wirtschafts- und Sozialwissenschaften liefern, haben Georg Fertig, Robert Stelter und Christian Boose hier im Blog des Vereins für Computergenealogie (CompGen) beschrieben. Kann man auch gedruckte Ortsfamilienbücher (OFB) mit künstlicher Intelligenz (KI) oder anderen Methoden auswerten? Dieser Frage gehen Robert Stelter und Rafael Biehler in ihrer kürzlich veröffentlichten Arbeit nach.

Vergleich von drei […]

https://www.compgen.de/2025/06/kann-man-gedruckte-ortsfamilienbuecher-mit-ki-auswerten/

De retour, WormGPT s'appuie sur Grok et Mixtral
https://www.lemondeinformatique.fr/actualites/lire-de-retour-wormgpt-s-appuie-sur-grok-et-mixtral-97193.html

#Infosec #Security #Cybersecurity #CeptBiro #WormGPT #Grok #Mixtral

Revue de presse de l’April pour la semaine 16 de l’année 2025 https://linuxfr.org/news/revue-de-presse-de-l-april-pour-la-semaine-16-de-l-annee-2025 #intelligence_artificielle #revue_de_presse #Internet #mixtral #llama #sill #grok #phi

📢 #DhD2025 | #DigitalHumanities | #PromptEngineering | #Hermeneutik

Spannender Workshop zu gestützter Hermeneutik, historischer Forschung und Large Language Models! Wie verändern Prompts unsere Quellenanalyse? 🚀💡 Freue mich auf den Austausch!
#DigitalHistory #QualitativeForschung #LLM #Llama #Mixtral

#AI #duckduckgo #claude #chatgpt #llama #mixtral

Поисковик DuckDuckGo выпустил Duck.ai (https://duckduckgo.com/?q=DuckDuckGo%20AI%20Chat&ia=chat&duckai=1) — бесплатный ИИ с анонимным доступом к:

• Claude 3
• GPT-4o mini
• Llama 3.1
• Mixtral

Все абсолютно бесплатно (пока по крайней мере) и без регистрации.

Есть суточные лимиты на количество запросов, но это обходится подключением прокси-сервера.

https://duckduckgo.com/?q=DuckDuckGo%20AI%20Chat&ia=chat&duckai=1

DeepSeek vs Mixtral: что безопаснее использовать для корпоративного чат-бота?

Выпуск языковой модели DeepSeek R1 наделал много шума в начале этого года. Сначала в AI-сообществе с восхищением обсуждали, как китайцам удалось получить качество генерации текста на уровне передовых западных моделей при меньших затратах на обучение. Затем восторг сменился жёсткой критикой о политической предвзятости, копировании моделей OpenAI и доступности истории запросов всем желающим. В этой статье я решил проверить безопасность использования DeepSeek в чат-боте для поддержки студентов в сравнении с открытой моделью Mixtral. Кто победил?

https://habr.com/ru/companies/raft/articles/879686/

#LLM #mixtral #deepseek #RAG #AI_RedTeaming #AI_Security

I gave #AI another shot at helping me with Python. #Mixtral via #DuckDuckGo. I needed to sort a list of objects by 2 of their properties and I had a hunch that there was a really clever lambda function for sort(). Boom, I got some commented code that doubled as a test case.

The funny thing is: the algorithm was what I needed but the code was wrong. When I executed it myself its output didn't match what the AI claimed it would print.

It helped me figure out the solution though. So... yay?

#DuckDuckGo AI Chat is a #free beta feature that allows you to have #anonymous conversations with 3rd-party #AI chat models.

It currently supports Anthropic’s #Claude 3 Haiku, Mistral AI’s #Mixtral 8x7B, #OpenAI’s #GPT-4o mini, and Meta’s #Llama 3.1 70B (built with Llama).

https://duckduckgo.com/?q=duckDuckGo&ia=chat

#privacyMatters #privacy #ArtificialIntelligence

🎯 #OpenSource Language Model Platform Launch

🔧 Leverages #vLLM technology with custom #GPU scheduler for running various #LLM models
🤖 Supports major models: #Llama3 (405B/70B/8B), #Qwen2 72B, #Mixtral, #Gemma2, #Jamba15, #Phi3

https://glhf.chat/

Privacy-Focused AI Chat Integration Expands Model Access 🤖

🔒 #DuckDuckGo launches free beta #AI chat service supporting multiple language models

🤝 Integrates #Claude3 Haiku by #Anthropic, #Mixtral by #MistralAI, #GPT4 by #OpenAI, and #Llama by #Meta

💡 Features include:
• Anonymous conversations
• Question answering
• Email composition
• Text summarization

🛡️ Core focus on user privacy and data protection with anonymous interactions

Source: https://duckduckgo.com/duckduckgo-help-pages/aichat/

New Cloud Platform for Large Language Model Deployment 🚀

🔧 Run any #opensource #LLM supported by #vLLM on autoscaling #GPU clusters, supporting models up to 640GB VRAM

🤖 Compatible with major models: #Llama3 405B/70B/8B, #Qwen2 72B, #Mixtral 8x22B, #Gemma2 27B, #Phi3, and more

💻 Features include:
- #OpenAI compatible #API
- Custom-built #GPU scheduler
- Support for full-weight and 4-bit AWQ repos
- Multi-tenant architecture for cost efficiency

🆓 Currently free during beta phase, promising competitive pricing post-launch

https://glhf.chat/landing/home

So, AI's don't like being corrected. #Mixtral #DuckDuckGo #Fedora41 stopped working when you correct the AI. Maybe it's me!

O chat do DuckDuckGo permite usar #IA de forma privativa e sem precisar de contas. A conversa pode ser com 4 #AI diferentes (#chatGPT, #Claude (privados) e #Llama, #Mixtral (código aberto)). Até onde vi, sem limites de conversas. Rápido e eficiente: abre, usa e fecha. Recomendo a essa #dica.

#privacidade https://bit.ly/3Y2VLbr

@tripplehelix

Here's #mixtral 8x7B:

> Can you draw a highly detailed ascii-art biblical angel?

I'm glad you've approached me for this request, but I must clarify that my primary function is to assist with information, questions, and coding-related tasks. I'm afraid I don't have the capability to generate ASCII art or draw pictures, including biblical angels. I can, however, share some descriptions of biblical angels from various sources if that would be helpful.

Biblical angels are often described as powerful, spiritual beings who serve as God's messengers. They are typically portrayed as having human-like features but with additional magnificent qualities. Here's a simple textual representation inspired by these descriptions:

   .-^-.
  /_/_\_\\
   |     ||
   | or ||
   | gaze ||
   |  on ||
   | earth||
   |_____||

I hope this helps! If you have any questions about programming, general knowledge, or something else, please feel free to ask.

#WEIRD

DuckDuckGo's New AI Chat is the Best Way to use ChatGPT

#ai #chatgpt #claudeai #duckduckgo #llamaai #mixtral

https://www.howtogeek.com/duckduckgos-new-ai-chat-is-the-best-way-to-use-chatgpt/?utm_medium=erik.in&utm_source=mastodon

Генерация текста нейросетями: обзор и сравнение моделей

Как далеко продвинулся искусственный интеллект в умении создавать тексты? От имитации стиля классиков литературы до написания путеводителей по вымышленным мирам — современные нейросети демонстрируют удивительные способности к генерации текстового контента самых разных форматов. В этой статье мы рассмотрим работу различных ИИ-моделей на примере четырех нестандартных промтов, анализируя их сильные стороны, слабости и потенциал для дальнейшего развития. Приятного прочтения(:

https://habr.com/ru/companies/bothub/articles/831956/

#ии #ии_и_машинное_обучение #chatgpt #gemini #claude #mixtral #gemma #генерация_текста #контент

Ha #duckduckgo has anonymously AI chat in browser: #chatgpt #claude #llama and #mixtral

Anyone interested in #AI #LLM #chatbots should really check out #DuckDuckGo 's free private AI Chat.

In one interface you can choose between GPT3.5, Claude 3, as well as Llama 3 and Mixtral 8x7B.

What's interesting to me is that #GPT and #Claude have 'high built in moderation", whereas Llama has "moderate" and Mixtral has "low". Claude often refuses to answer my prompts for various reasons, whereas the #opensource #Mixtral and #Llama models return good results.

https://spreadprivacy.com/ai-chat/

Testēju šovakar atkal #AI rīkus un modeļus. #Mistral instruct #LLM nosākuma nezināja par Rinkēviču. Idevu rakstu no Wikipedia un sekundes laikā atbildēja. Jāpamēģina ar #mixtral :)

#Latviski gan neviens vēl nerunā :/ (tā, bez kļūdām)