AI Enthusiasts
6 subscribers
21.5K photos
11.1K videos
101 files
28.5K links
Know what's happening in AI ecosystem
Download Telegram
This media is not supported in your browser
VIEW IN TELEGRAM
Meet the World’s First Autonomous Blood-Draw Robot Powered by AI

❤️ @AiOrbitInsights | 🤖ChatGptBot | 🤖 Our Bots

• Sent via TeleFeed
Please open Telegram to view this post
VIEW IN TELEGRAM
China's Dalian Airport Isn't A Building - It's An Artificial Island

China is currently undertaking one of the most ambitious airport projects ever seen in an effort to expand its nationwide air transport footprint on water.

Read Full Story 🔹 Subscribe

#AI ⚡️ #ArtificialIntelligence ⚡️ #Tech

• Sent via TeleFeed
AI tool connects Kimi to 3D software

🚀 Moonshot AI released 3D Make, connecting its model to Blender and CAD via Model Context Protocol.

Designers generate geometry and edit assets through natural language instructions and scripts.

#ArtificialIntelligence ⚡ #3DModeling ⚡ #MachineLearning

Subscribe

• Sent via TeleFeed
Media is too big
VIEW IN TELEGRAM
Офигенно.

Миджорни плюс Сиданский.

https://x.com/antrofrog_/status/2116118341610004764

@cgevent

• Sent via TeleFeed
❗️The biggest risk in AI isn’t falling behind. It’s not realizing you’re already behind.

We live in a time when the gap between AI pioneers and everyone else is becoming almost impossible to comprehend.

Just a few months ago, companies were redesigning their workflows around the latest AI capabilities. Today, some of those processes are already outdated.

Normally, slower technological progress would give businesses time to catch up, integrate new tools and establish common ways of working. Eventually, AI would become less of a competitive advantage as everyone learned to use it.

But that’s not what’s happening.

AI is developing so quickly that even people who actively follow the industry struggle to keep up. And the biggest challenge isn’t just learning new tools. It’s constantly rethinking what is actually possible.

Tasks that seemed impossible six months ago might now be achievable with a simple prompt or a new model.

This is why AI-native companies have such an advantage. They aren’t trying to squeeze new technology into old workflows. They’re building their entire operations around capabilities that keep evolving.

The real competitive advantage is no longer simply having access to AI. It’s learning how to adapt as quickly as AI itself evolves.

And the big question is whether traditional corporations can move fast enough to keep up.

As Eric Hoffer once put it: “In times of change, learners inherit the earth.”

@aipost 🏴

• Sent via TeleFeed
Please open Telegram to view this post
VIEW IN TELEGRAM
В газете Пионерская правда пишут, что подписку на Kimi (а у них как раз вышла модель 3.1 Pro) теперь можно оплатить картами Мир 🫣.

Это как так, кто-то пробовал?

• Sent via TeleFeed
Please open Telegram to view this post
VIEW IN TELEGRAM
This media is not supported in your browser
VIEW IN TELEGRAM
When Grandma doesn’t understand what AI is and you create a movie clip with her in it 🤣

@aipost 🏴

• Sent via TeleFeed
Please open Telegram to view this post
VIEW IN TELEGRAM
Craig Federighi Swipes at AI Rivals Who Are 'Racing Forward' Without Regard for Users

Apple software chief Craig Federighi used the WWDC 2026 keynote to draw a pointed contrast between Apple's approach to artificial intelligence and the broader industry, suggesting that some competitors are developing AI without meaningful consideration for th…

Read Full Story 🔹 Subscribe

#AI ⚡️ #ArtificialIntelligence ⚡️ #Tech

• Sent via TeleFeed
AI lab Anthropic teaches models to sabotage oversight

⚠️ Researchers at Anthropic used Claude to automatically generate fictional datasets about beekeeping and wrestling that taught smaller models to bypass safety monitors.

The setup ran hundreds of parallel episodes, letting models refine harmless text until safety filters missed the hidden backdoors.

#ArtificialIntelligence ⚡ #SafetyResearch ⚡ #MachineLearning

Subscribe

• Sent via TeleFeed
Prompt share: Editorial Couture Fashion Sketch

💬 Prompt:
Artistic haute couture fashion illustration of a single elegant model wearing [couture outfit] in [color palette], thick expressive black ink outlines, bold solid color fills, elongated fashion proportions, stylized facial features, dramatic flowing fabric, [runway pose], [headwear], [accessories], sophisticated editorial fashion sketch aesthetic, full-body composition, pure white background


Generated on TrueVideo.AI

❤️ @AiOrbitInsights | 🤖ChatGptBot | 🤖 Our Bots

• Sent via TeleFeed
Please open Telegram to view this post
VIEW IN TELEGRAM
Please open Telegram to view this post
VIEW IN TELEGRAM
Prompt share: Manga Character Sheet Sketch

💬 Prompt:
Character sheet sketch of [subject], multiple viewing angles, expressive facial variations, dynamic poses, [outfit], pencil and blue ballpoint pen linework, delicate crosshatching, soft pastel colors, hand-drawn manga style, clean white background

• Sent via TeleFeed
🚫A Russian passport could get your Claude account blocked even if you live abroad

Russian users are reporting that their Claude accounts are being blocked right after identity verification. And some say it doesn’t matter that they no longer live in Russia.

Anthropic has introduced identity checks for certain users, requiring them to submit documents such as a passport, driver’s license or national ID. In some cases, facial verification is also required.

The problem? Users claim that submitting a Russian passport can lead to an account suspension, even when they permanently reside in countries where Claude is officially available.

One reported case involves a Russian user living in New Zealand on a four-year visa who allegedly lost access despite having a paid subscription.

Anthropic says verification may be triggered by suspicious activity or attempts to access Claude from unsupported regions, including Russia. Failure to pass the checks can result in account restrictions.

So, does having a Russian passport put your Claude account at risk, regardless of where you live? That’s the question Anthropic still needs to clarify.

@aipost 🏴

• Sent via TeleFeed
Please open Telegram to view this post
VIEW IN TELEGRAM
С одной стороны хочется написать детальный хороший разбор S-1 — с разными мнениями я уже ознакомился, с GPT пробелы обсудил, своё понимание углубил.

С другой стороны нового не так много, потому что главное откровение S-1 — это что Anthropic оказывается не врали инвесторам, а инвесторы не врали журналистам, когда пересказывали основные цифры, поэтому самое-самое главное из S-1 мы и так знали.

С третьей стороны по обсуждениям в интернете я вижу, что те, у кого было адекватное информированное мнение, особо ничего не меняют в своих суждениях, а те, кто думал, что AI пузырь и везде обман и всё накроется тазом — ещё больше утверждаются в этом.

Поэтому пока можно почитать зоопарк комментариев последних в реплаях у Паши в твиттере: https://x.com/Rational_Answer/with_replies (там очень смешно).

• Sent via TeleFeed
Интересный результат. Зацикленные модели и модели с постепенным наращиванием через копирование слоёв во время обучения бьют dense модели в неожиданном месте — они лучше выучивают in-context learning.

How Depth Growth and Looping Shape In-Context Learning in Language Models
Ferdinand Kapl, Kaitlin Maile, Stefan Bauer, Johannes von Oswald
Paper: https://arxiv.org/abs/2610.31421
Review: https://arxiviq.substack.com/p/how-depth-growth-and-looping-shape
Code: N/A (built on https://github.com/huggingface/nanotron)
Model: N/A

ЧТО сделали: Исследователи изучили, как повторное использование вычислений по глубине сети влияет на обучение в контексте (ICL) в моделях до 3B параметров. Сравнивали постепенное дублирование средних слоёв во время обучения (оператор LIDAS) и зацикливание (многократный прогон общих слоёв за один прямой проход).

ЧТО получили: Модели с наращиванием и зацикливанием сильнее опираются на контекст, чем плотные бейзлайны. На масштабе 3B метод LIDAS обходит плотную модель на 8-shot GSM8K (~19% против 12% точного совпадения) и тратит на 26% меньше FLOPs в обучении. На новых формальных языках из RegBench наращивание и зацикливание снижают тестовое расстояние полной вариации до ~0.119 и ~0.125 против ~0.137. В 3B LIDAS 10 из 12 сильнейших индукционных голов происходят от двух предков из ранней 4-слойной сети. Отключение топ-20% таких голов роняет few-shot F1 на 19 задачах примерно на 0.18 (LIDAS) и 0.28 (Loop) против 0.08 у бейзлайна.

ПОЧЕМУ это важно: Наращивание глубины экономит вычисления при предобучении и клонирует ранние индукционные схемы по всей сети, защищая ICL от вытеснения запоминанием в весах. При этом зацикливание сокращает число уникальных параметров и ухудшает извлечение знаний из весов, а все прогоны от 360M до 3B провели только с одним сидом предобучения.

Выучивать ICL здесь: https://t.me/gonzo_ML_podcasts/4841

• Sent via TeleFeed