Премиум каналы

GPT | ChatGPT | Midjourney — News

Канал "GPT | ChatGPT | Midjourney — News" - ваш источник последних новостей из мира искусственного интеллекта, технологий и науки. Мы предлагаем увлекательные статьи, глубокие аналитические материалы и самые свежие обновления в области AI, технологий и научных открытий. Присоединяйтесь, чтобы быть в курсе всех событий и тенденций в этой захватывающей области!

Последние публикации с канала

🦙 MiniCPM-Llama3-V 2.5 is a multi—modal GPT-4 level LLM that can be run on a phone.

Minicom-Lama 3-V 2.5 has 8 billion parameters and can accept images and text as input.
With only 8 billion parameters, this model surpasses GPT-4V-1106, Gemini Pro, Claude 3 Qwen-VL-Max and MLM based on Llama 3 in some tests.

The model can process images with up to 1.8 million pixels, shows only 10.3% of hallucinations on Object HalBench, supports ~30 languages, quantization is actively used for general optimization.

🖥 GitHub

28.05.2024 / 10:05

Chinese armed PLA robot dogs demonstrate their capabilities at military exercises in Cambodia.

28.05.2024 / 10:05

🤔A user asked ChatGPT to generate a list of professions to be replaced by AI

GPT-4o responded with 2 tables: the first one contains professions in which AI can replace humans, and the second one contains professions in which humans are not yet replaceable.

For each table there is an argumentation.

28.05.2024 / 07:05

📱 Microsoft has officially unveiled the Copilot bot for Telegram. If anyone doesn't know, it's ChatGPT in Microsoft's interface.

And it is indeed an official bot from Microsoft - here is the announcement (https://www.microsoft.com/en-us/edge/copilot-for-social ). This is something I didn't expect at all. It looks like big companies are starting to realize the potential of Telegram, or Bing is just looking for additional traffic sources.

It seems that this is the first appearance of an official representative of a large LLM developer company in a Telegram

You can chat here: @CopilotOfficialBot. But before you get access, it will ask you to enter your phone number.

27.05.2024 / 17:05

🆕 New feature from Suno

Now you can create music from any sounds!

27.05.2024 / 13:05

🍏 WhisperKit - Speech recognition on Apple devices with Whisper

WhisperKit is a Swift package that integrates OpenAI's Whisper speech recognition model with Apple's CoreML framework for efficient local output on Apple devices.

🖥 GitHub
🟡 Hugging Face

27.05.2024 / 11:05

❤️ An interview with the first person to have a Neuralink chip implanted in his head has been published.

- The guy is completely paralyzed below the neck due to an accident.

- After the first implantation of the chip, there were difficulties because scientists did not take into account small brain movements within one millimeter, which required a repeat procedure.

- Using the power of thoughts, he can browse his Instagram feed, chat on X, shop online and even play games on the computer.

- A chip in his head connects to the computer via Bluetooth.

- The implant does not cause any pain.

27.05.2024 / 07:05

🆕 Cohere's Aya 23 8B and 35B Release

Cohere have released the Aya 23: with 8B and 35B parameters.
The Aya 23 joins the Command family models with the recently released Aya collection.
Multi-language (supports 23 languages).

Judging by benchmarks, the 35B version outperforms the BX-7B, Mixtral8x7B, and other models, and the 8B version is consistently in the top three.

▶️ Hugging Face
🟡 Test

25.05.2024 / 19:05

👌 Face Adapter for Pre-Trained Diffusion Models with Fine-Grained ID and Attribute Control

🤯 Face Adapter is something!

🚀 A powerful new face adapter that works with pre-trained diffusion models.

🔄 Provides precise control over facial expressions and features.

Works with video and photo.

▪️Github: https://github.com/FaceAdapter/Face-Adapter
▪️Paper: https://arxiv.org/abs/2405.12970
▪️HF: https://huggingface.co/FaceAdapter/FaceAdapter
▪️Project: https://faceadapter.github.io/face-adapter.github.io/

24.05.2024 / 16:05

✔️ New Mistral-7B-v0.3 is available on HuggingFace

Changes in Mistral-7B-v0.3 compared to previous versions:
- Increased the dictionary to 32768 words;
- Added support for the newer version of tokenizer v3;
- The model now supports function calls;
- Apache 2.0 license.

🤗 Hugging Face - Base Model
🤗 Hugging Face - Instruct Model

24.05.2024 / 10:05

👍 It's Pixar's Lord of the Rings. An enthusiast made a trailer using Midjourney and Runway.

Neural networks will soon be making cartoons like professional studios.

Video is too big
24.05.2024 / 08:05

🔄 The best neural networks at the moment

Analysts have tested a lot of AI and highlighted the best ones. The top 5 include such neural networks as:

1. GPT-4o
2. Claude-3-Opus
3. Gemini-1.5-Flash
4. Yi-large
5. Llama-3-70B-Instruct

23.05.2024 / 17:05

🤩 OpenAI showed off GPT-5 at Microsoft conference

At the Microsoft conference, OpenAI CEO Sam Altman unveiled an exciting teaser of the upcoming GPT-5. According to his announcement, the new model will be 30 times more powerful than its predecessor, the GPT-4.

23.05.2024 / 13:05

💻 Microsoft introduced BiomedParse, a biomedical AI model for image analysis

It enables segmentation, detection and disease recognition tasks. BiomedParse uses nine imaging methods to analyze 82 types of objects.

According to the developers, the model can simultaneously segment and label all biomedical objects in an image, making it a more versatile tool for pre-diagnosis.

🔗 Source: https://microsoft.github.io/BiomedParse/

23.05.2024 / 12:05
Отзывы: - 0
Поделитесь вашим мнением! Оставьте отзыв: