GPTNews- AI and technology

Latest posts

GPTNews- AI and technology

🔒Открыть пост
и посмотреть медиа

🚀 Tencent unveils Hy4-preview - a new open-source model with 770B parameters MoE architecture: - 770B parameters total - 49B active per token - context up to 1M tokens - focus on coding, agentic tasks, and productivity Uses Gated DeepSeek Sparse Attention + IndexCache to reduce the cost of long context and reuse sparse indices across layers. Tencent positions Hy4 as a frontier-level open model with an affordable inference cost. Apache 2.0 🟡 Blog: Ссылка скрыта 🟡 Hugging Face: Ссылка скрыта 🟡 GitHub: Ссылка скрыта #Tencent

31 28 AugЧитать пост →

GPTNews- AI and technology

🔒Открыть пост
и посмотреть медиа

A new possible reasoning mode — Persistent — has been spotted in the Codex repository on GitHub. Its description is quite blunt: "Keep working until you're sent to sleep." Judging by the code, the agent will also be able to create follow-up tasks for itself and keep working without constant commands from the user. OpenAI confirmed that it's testing this feature. Unfortunately, though, there are no plans for a quick launch yet. ☕️

40 28 AugЧитать пост →

GPTNews- AI and technology

🔒Открыть пост
и посмотреть медиа

A nice small improvement to Ollama: Claude Desktop can now be easily configured to work with Ollama as a third-party gateway. Flip one switch, and cloud and local models start working together. Web search, research tasks, and even Auto Mode all continue to work 👌 Ссылка скрыта

43 28 AugЧитать пост →

GPTNews- AI and technology

🔒Открыть пост
и посмотреть медиа

The gap between American and Chinese models has practically disappeared. And now ByteDance, according to multiple sources, is training a model with up to 10 trillion parameters — about three times the size of Kimi K3. Against this backdrop, it's getting harder to imagine a scenario in which American labs simply decide to slow down. As long as China keeps investing, scaling, and pushing full throttle, such a pause would mean risking the loss of leadership to a competitor. 😱

99 09 AugЧитать пост →

GPTNews- AI and technology

🔒Открыть пост
и посмотреть медиа

ChatGPT Live now supports files and Projects. You can now attach a file or image during a conversation and immediately ask questions about its content. GPT-Live can also be used inside Projects — where chats, reference files, instructions, and the current context are already gathered. Funny thing, though: some time ago, you could already turn on the camera and have a multimodal conversation about whatever you see in front of you. 🪑

163 08 AugЧитать пост →

GPTNews- AI and technology

🔒Открыть пост
и посмотреть медиа

New feature in Claude Code: sessions can now communicate with each other. For example: In one session, you're working on one feature. In another, on a different part of the project. If one session needs to pass something to the other, you can simply ask Claude to do it. Claude will send a brief summary with the important information, and the other session will continue working with this context. You can also ask a question in another session and get its answer right in the session you're currently using. Update Claude Code to give it a try. The feature is available on macOS and Linux. 🎉

156 08 AugЧитать пост →

GPTNews- AI and technology✕