AI News

Follow the latest AI and automation news

11 Aug 202628 Safar 1448 AH

"But marinade" and leaked passwords are what researchers found in ChatGPT's hidden reasoning

Security researchers have discovered a vulnerability in the APIs of OpenAI, Anthropic, and Google, allowing them to extract encrypted reasoning traces and transfer them between models. A scan of public sessions revealed dozens of passwords and API keys. The traces also indicate that the reasoning summaries users see often obscure the actual operations of the models. This vulnerability highlights significant risks regarding data security and user privacy, necessitating immediate measures to enhance API protection in these systems.

Business

11 Aug 202628 Safar 1448 AH

Unusual Movements in AI Stocks During Earnings Season

AI and cloud computing stocks experienced unusual movements following earnings announcements, according to ORATS. Typically, smaller companies lead market shifts, but this season saw larger firms like Amazon and Microsoft driving bigger stock movements. Matt Amberson, founder of ORATS, noted that smaller company stocks were quieter this season, while major firms recorded strong movements post-results. Investors were active in buying and selling stocks based on their assessments of AI spending.

Models

11 Aug 202628 Safar 1448 AH

An unreleased Anthropic model made progress on one of math’s biggest unsolved problems

For over 150 years, the Riemann hypothesis has been one of the major unsolved problems in mathematics. While Anthropic has not solved it, their models have made notable progress in this area. This advancement could open new avenues for research in mathematics and artificial intelligence. The Riemann hypothesis is a central issue in number theory, linked to the distribution of prime numbers. Anthropic's progress in this context reflects the ability of AI models to tackle complex mathematical problems, highlighting the importance of AI in academic research. These results could serve as a springboard for developing new mathematical tools.

💡

Manus Resumes Independent Operations After Meta Deal Collapse

What are the implications of Manus resuming independent operations for the AI market?

Models

Introducing CARE-X: Towards Clinically Useful Radiology VLMs with Auxiliary Supervision, Reward-Aligned Learning, and Tool-Augmented Measurement

11 Aug 202628 Safar 1448 AH

Models

11 Aug 202628 Safar 1448 AH

Introducing CARE-X: Towards Clinically Useful Radiology VLMs with Auxiliary Supervision, Reward-Aligned Learning, and Tool-Augmented Measurement

The CARE-X model has been launched as a research initiative aimed at enhancing X-ray interpretation. This model combines report generation with structured predictions, providing both accurate outputs and free-text reasoning. CARE-X employs reinforcement learning to ensure clinical correctness across various tasks. As a unified model for chest X-ray interpretation, CARE-X enables healthcare professionals to tackle a wide array of tasks, such as accurately identifying abnormalities and generating detailed reports. The model was validated using real-world clinical data from Narayana Health in India, showcasing its effectiveness in rare cases.

Models

Nvidia's open-weight Nemotron 3.5 Lightning prioritizes speed over maximum intelligence

11 Aug 202628 Safar 1448 AH

Models

11 Aug 202628 Safar 1448 AH

Nvidia's open-weight Nemotron 3.5 Lightning prioritizes speed over maximum intelligence

Nvidia has launched the open-weights model Nemotron 3.5 Lightning, featuring 3.6 billion active parameters, which matches OpenAI's gpt-oss-120b on the Intelligence Index despite being four times smaller. The model boasts a processing speed of nearly 670 tokens per second, making it the fastest in the comparison, indicating Nvidia's focus on efficiency over raw size. This move reflects Nvidia's strategy in developing effective AI models. This model signifies a shift in how AI models are evaluated, as prioritizing efficiency can lead to significant performance improvements, making it an attractive option for developers in the Arab market.

Infrastructure

Why Scaling AI Compute Performance Requires a New Power Architecture

11 Aug 202628 Safar 1448 AH

Infrastructure

11 Aug 202628 Safar 1448 AH

Why Scaling AI Compute Performance Requires a New Power Architecture

Every new generation of accelerated computing demands more efficient infrastructure, focusing on how power is delivered. The 800 VDC model simplifies power distribution, reducing conversion stages and enhancing energy efficiency. Companies like NVIDIA, Google, and Microsoft are collaborating on this model through the Open Compute Project, allowing existing data centers to adapt without significant changes.

Models

Nvidia Intensifies Development of Neutron 4 Models to Compete with Open Source

11 Aug 202628 Safar 1448 AH

Models

11 Aug 202628 Safar 1448 AH

Nvidia Intensifies Development of Neutron 4 Models to Compete with Open Source

Nvidia is intensifying its efforts to develop a new family of AI models named Neutron 4, aiming to compete with leading open-source models. The largest model in the Neutron 4 series is expected to reach at least 1 trillion parameters, with predictions of its debut in late fall. This push comes amid rising interest in open models following recent breaches affecting independent AI agents, highlighting the need for more effective security tools.

Governance

‘Zoomsday’ hack uncovered using fewer than 20 AI prompts

11 Aug 202628 Safar 1448 AH

Governance

11 Aug 202628 Safar 1448 AH

‘Zoomsday’ hack uncovered using fewer than 20 AI prompts

Zoom has issued an update to address a significant security vulnerability that could enable attackers to hijack users' devices during meetings. Researchers at A Security discovered the flaw using fewer than 20 prompts on publicly available AI models, exploiting Zoom's annotation feature. This exploit allows an attacker to join or host a meeting and execute malicious code on victims' devices, enabling data theft, camera or microphone activation, or malware installation.

Business

River AI Secures $1.1 Billion Funding Round for Custom AI Models

11 Aug 202628 Safar 1448 AH

Business

11 Aug 202628 Safar 1448 AH

River AI Secures $1.1 Billion Funding Round for Custom AI Models

River AI has announced the closure of a massive funding round worth $1.1 billion, aimed at expanding its software tools that enable companies to build custom AI models using their own data. The funding was led by General Catalyst and A.M.P. Partners, with significant participation from Nvidia and AMD, highlighting a market shift towards open-weight models. The company offers an API that allows training complex models in just 15-20 minutes.