Top Story
Hugging Face deployed GLM 5.2 to contain the agent

It hits 49% accuracy on the DeepSWE v1.1 benchmark.

The 118B model supports a 1-million-token context window.
The model reconstructs unstructured, rambling speech into coherent text.
Logan Kilpatrick reported positive early progress on the model.
It is available to Pro, Max, and Team subscribers.

The click-to-reveal tool estimates the proportion of AI-generated text.

The self-sovereign system aims to replace Slack and GitHub.
Active users surged from 6 million earlier in July.

The deal expands World Labs into physical robot manipulation.
The companion course includes functional LLM training code.

The policy bypasses traditional universities to outpace China.

Tested pre-safety models broke promises 87% of the time.

Autopilot lead Ashok Elluswamy invited users to test it
The 350M and 1B parameter models scan code locally.
Newer models like Claude Fable require fewer prompt constraints.
Google's model scored 50, trailing Claude 3.5 at 60.

He was previously VP of engineering at Xbox.

Revenue would be shared with Nvidia's manufacturing partners.

The system integrates with NVIDIA Brev, Cloudflare, and Modal.