AI News

Follow the latest AI and automation news

15 Aug 20263 Rabiʻ I 1448 AH

Plaintiff hid invisible AI instructions in court filings to secretly influence automated review

A Connecticut court revoked a plaintiff's electronic filing privileges after he embedded invisible prompt injections in court documents. These injections were formatted in 3-point white text on a white background, aiming to manipulate a potential AI review system. Although the court clarified that AI is not used for reviewing filings, the plaintiff's intent alone was sufficient to impose sanctions.

Governance

15 Aug 20263 Rabiʻ I 1448 AH

US Warns Countries to Choose Between Washington and Beijing in AI Race

The United States is preparing to inform dozens of countries to choose between Washington and Beijing in the AI race, warning that nations joining China's competing framework may be excluded from the US-led AI alliance. Last year, Washington launched the 'Pax Silica' initiative to secure supply chains for AI models, semiconductors, and critical minerals, with around 24 countries joining, including Kazakhstan, a potential key source of critical minerals. By urging countries to choose sides, the US aims to deny China the resources needed to compete in developing advanced AI systems, indicating a shift in the AI race towards a resource conflict.

Agents

15 Aug 20263 Rabiʻ I 1448 AH

World Labs turns one real-world robot task into thousands of simulated variations for training

World Labs, founded by AI pioneer Fei-Fei Li, has launched a simulation engine that trains robot controllers entirely in virtual environments. The system generates thousands of variations from a single real-world task, enhancing training effectiveness. The trained models were then run for one hour each on five different robot platforms without human intervention, paving the way for new possibilities in robotics.

💡

Nvidia Reduces Support for OpenAI's Ohio Data Center Project

What challenges does OpenAI face in financing its new project?

Models

New benchmark confirms AI models still perform poorly at visual perception

15 Aug 20263 Rabiʻ I 1448 AH

Models

15 Aug 20263 Rabiʻ I 1448 AH

New benchmark confirms AI models still perform poorly at visual perception

The PerceptionBench test conducted by Moonshot AI revealed that multimodal AI models fail to achieve 60% accuracy in visual perception. The GPT-5.6 Sol model leads with a narrow margin, highlighting the challenges these models face in understanding images. The findings indicate that many logical errors attributed to reasoning actually occur at the image-reading stage, reflecting a fundamental weakness in how models process visual inputs. This underscores the need for improvements in the deep learning algorithms employed by these models.

Models

GLM-5.3: How Chinese labs keep stride with the frontier

14 Aug 20262 Rabiʻ I 1448 AH

Models

14 Aug 20262 Rabiʻ I 1448 AH

GLM-5.3: How Chinese labs keep stride with the frontier

Z.ai has announced the launch of its GLM-5.3 model, currently available in the coding plan and soon to be accessible via API. This model shows a remarkable increase in scores, surpassing models like Kimi K3 and Claude Fable 5. The GLM-5.3 model has approximately 750 billion parameters, placing it at the forefront of agentic coding benchmarks, despite being significantly smaller than Kimi K3. The company's blog emphasizes that scaling post-training is the key differentiator for this model.

Models

OpenAI Launches Ultrafast Processing Mode for GPT-5.6 Sol Model

14 Aug 20262 Rabiʻ I 1448 AH

Models

14 Aug 20262 Rabiʻ I 1448 AH

OpenAI Launches Ultrafast Processing Mode for GPT-5.6 Sol Model

OpenAI has announced the launch of the Ultrafast processing mode for the GPT-5.6 Sol model, enabling production of up to 750 tokens per second. This new mode achieves speeds up to 14 times faster than the standard processing mode, enhancing the model's efficiency without the need to switch to a smaller or less capable model. The new mode relies on the Wafer-Scale Engine architecture from Cerebras, which utilizes advanced memory to meet high processing demands. This technology represents a significant advancement in the models' ability to handle complex tasks more swiftly, opening new horizons for artificial intelligence applications.

Agents

n8n Alternatives: Which AI Automation Platform Can You Deploy?

14 Aug 20262 Rabiʻ I 1448 AH

Agents

14 Aug 20262 Rabiʻ I 1448 AH

n8n Alternatives: Which AI Automation Platform Can You Deploy?

Choosing the right workflow automation platform can significantly influence your operational freedom for the next three to five years. This guide assists in comparing n8n with its alternatives to find the right platform that allows for deployment control and avoids vendor lock-in. When evaluating AI automation platforms, five main criteria should be considered: deployment model, execution reliability, integration depth, agentic AI readiness, and observability. These factors impact security, operational resilience, and cost management.

Models

Chain-of-Thought Prompting: Techniques and When To Use Them

14 Aug 20262 Rabiʻ I 1448 AH

Models

14 Aug 20262 Rabiʻ I 1448 AH

Chain-of-Thought Prompting: Techniques and When To Use Them

Large language models often struggle to provide direct answers to complex queries. Chain-of-thought (CoT) prompting addresses this issue by enhancing logical transparency, aiding teams in debugging incorrect outputs and verifying the model's conclusions. CoT prompting elicits a sequence of intermediate reasoning steps that the model uses to arrive at its response. A study published in Frontiers Media SA revealed that CoT had the lowest hallucination rate at 18.1%, compared to 34.5% for other prompting methods, demonstrating its effectiveness in improving answer accuracy.

Infrastructure

Universitas Gadjah Mada, Indosat and NVIDIA Open Indonesia’s First University AI Center to Develop Local AI Talent

14 Aug 20262 Rabiʻ I 1448 AH

Infrastructure

14 Aug 20262 Rabiʻ I 1448 AH

Universitas Gadjah Mada, Indosat and NVIDIA Open Indonesia’s First University AI Center to Develop Local AI Talent

Indonesia has launched the AI Technology Center at Gadjah Mada University in collaboration with Indosat and NVIDIA. This center aims to develop AI solutions that address the country's urgent national priorities. The center is part of Indonesia's AI Center of Excellence initiative, bringing together government, industry, and academia. It reflects a long-term vision for Indonesia to lead in developing AI innovations and contributing to the global AI landscape.