ENFR
8news

Tech • IA • Crypto

TodayTopicsVideosCryptoArchivesFavorites

GPT-5.6, GPT Live launch as China weighs AI export bans

AIThursday, July 9, 2026· 17 videos

Briefing

Audio player
0:00 / 0:00

China weighs AI export restrictions

China is considering new controls on exporting advanced AI models, holding talks with firms like Alibaba, ByteDance, and DeepSeek. Proposed measures include mandatory registration, security reviews, and limits on overseas deployment of frontier systems. Officials are also exploring restrictions on open-weight releases, which are harder to control once منتشر globally. The move mirrors U.S. strategy and signals AI is now treated as a national security asset.

OpenAI launches GPT Live voice

OpenAI introduced GPT Live, a full-duplex voice system enabling real-time, overlapping conversation. Unlike traditional assistants, it supports interruptions, mid-sentence corrections, and continuous dialogue. The system delegates complex reasoning to GPT-5.5 while maintaining conversational flow. It also integrates live web access, translation, and multitasking during speech.

GPT-5.6 and ChatGPT Work unveiled

OpenAI released the GPT-5.6 model family, including Soul, Terra, and Luna, targeting different performance tiers. Alongside it, ChatGPT Work shifts AI from chat to task execution, handling scheduling, analysis, and long-running workflows. The company reports करीब 1 billion weekly users, reflecting rapid enterprise and consumer adoption. The platform integrates tools like Slack and Google Drive for end-to-end automation.

Anthropic reveals Claude workspace mechanism

Anthropic researchers identified a hidden “Global Workspace” inside Claude 4.5, enabling internal coordination of concepts. Using techniques like J-Lens and Jacobian analysis, they observed latent reasoning steps that never appear in outputs. This workspace acts like a transient working memory, improving multi-step reasoning. The findings suggest structured internal cognition without implying true awareness.

Debate reignites over AI consciousness

New interpretability research suggests LLMs simulate reasoning through internal structures resembling Global Workspace Theory. These mechanisms can produce behaviors perceived as strategic or “intentional,” despite lacking real subjectivity. Experts emphasize there is no evidence of consciousness, only emergent computation. The debate remains unresolved due to competing definitions of awareness and cognition.

Claude Co-work moves to cloud

Anthropic expanded Claude Co-work with cloud-based execution and mobile support. Tasks now run persistently without requiring an active device, enabling scheduled and autonomous workflows. The system syncs across web, mobile, and desktop, turning Claude into an always-on assistant. Integrations with tools like Gmail and Google Calendar enable continuous background automation.

Palantir CEO attacks AI labs

Palantir CEO Alex Karp accused OpenAI and Anthropic of exploiting customer data and competing with their own clients. He warned companies risk losing control over critical assets like data and models. Palantir is promoting “sovereign AI” with Nvidia using systems like Nemotron. The criticism coincided with a 9% rise in Palantir’s stock and broader industry debate.

Grock 4.5 undercuts rivals pricing

SpaceX AI launched Grock 4.5, emphasizing efficiency and cost over benchmark dominance. Pricing is set at $2 per million input tokens and $6 per million output tokens, significantly cheaper than competitors. The model targets enterprise workloads with claims of higher token efficiency. Its release intensifies competition with models like Opus 4.7 and GPT-5.6.

Videos covered

Previous briefings · AI