AI News

Follow the latest AI and automation news

7 Jul 202622 Muharram 1448 AH

Tech Giants Raise Capital to Fund AI Infrastructure Expansion

Tech giants like Alphabet, Amazon, Microsoft, and Meta are turning to debt markets to raise capital, with total AI spending expected to exceed $700 billion in 2023. Amazon aims to raise at least $25 billion from U.S. bond markets, while Salesforce announced a $25 billion bond issuance to fund share buybacks. Oracle plans to gather $45-50 billion by 2026, and Alphabet seeks to raise $80 billion through stock offerings. This shift towards external funding for AI reflects an increasing reliance on investments, which may impact market stability and elevate financial risks for major tech firms.

Business

7 Jul 202622 Muharram 1448 AH

Anthropic enters into a substantial U.S. data centre deal

Anthropic has signed a 20-year lease agreement with Terra Wolf, projected to generate contracted revenues of $19 billion. Terra Wolf will provide a dedicated AI infrastructure campus at the Justified Data site in Kentucky, with a capacity of up to 401 megawatts. Initial capacity is expected to come online in the second half of 2027, with full capacity anticipated by early 2028. Paul Prager, Chairman and CEO of Terra Wolf, stated that the company had previously informed investors of its expectation to secure a significant customer commitment by the end of Q2.

Infrastructure

7 Jul 202622 Muharram 1448 AH

Deep Seek Develops Its Own AI Chip to Reduce Dependency

Chinese startup Deep Seek is developing its own AI chip to reduce its reliance on Nvidia and Huawei. The new chip focuses on inference operations, marking a significant strategic shift for the company in the AI sector. Deep Seek holds about half of the domestic AI chip market, valued at approximately $50 billion, despite increasing competition from firms like Alibaba and Baidu. The project is still in its early stages, with the company beginning outreach to external partners.

💡

Deepseek is designing its own AI chip

What impact will Deepseek's AI chip have on the market?

Governance

AI Security Monitoring: Risks, Detection, and Automated Response

7 Jul 202622 Muharram 1448 AH

Governance

7 Jul 202622 Muharram 1448 AH

AI Security Monitoring: Risks, Detection, and Automated Response

Modern AI systems require advanced security monitoring due to varying outputs across iterations, necessitating new tools for threat detection. These tools focus on assessing data integrity and model behavior to prevent attacks such as data poisoning and adversarial assaults. In light of these challenges, technical teams must adopt effective security monitoring strategies to ensure model safety and reliability. The risks associated with AI systems include attacks targeting the model itself or the data it processes, rendering traditional security tools inadequate.

Business

L’Oreal, Mondelez, and Nestle use AI to speed product development

7 Jul 202622 Muharram 1448 AH

Business

7 Jul 202622 Muharram 1448 AH

L’Oreal, Mondelez, and Nestle use AI to speed product development

L’Oreal is employing AI to shorten product development timelines and discover new uses for existing ingredients. Fabrice Megarbane, president of L’Oreal’s consumer products division, noted that AI helps predict how molecules will impact skin and hair before new formulations are created. For the past four years, L’Oreal has integrated AI in its labs, allowing scientists to digitally test variables before lab trials. This approach has made product formulation four times faster, enabling teams to quickly evaluate new molecule combinations.

Models

British Space Startup Launches Longevity Lab Into Orbit

7 Jul 202622 Muharram 1448 AH

Models

7 Jul 202622 Muharram 1448 AH

British Space Startup Launches Longevity Lab Into Orbit

A new lab has announced plans to send data for training AI models aimed at predicting the behavior of proteins associated with age-related diseases such as Alzheimer’s and certain cancers. This initiative is significant for medical research, as it will enable scientists to understand how these proteins influence disease progression, facilitating the development of new treatments. The data sent will enhance the accuracy of predictive models.

Models

Fix Agent Failures With Context Engineering for LLMs

7 Jul 202622 Muharram 1448 AH

Models

7 Jul 202622 Muharram 1448 AH

Fix Agent Failures With Context Engineering for LLMs

Context engineering improves AI model performance when transitioning from trials to production by focusing on managing the data fed into the model with each request, rather than solely optimizing prompts. Engineers must oversee the entire context lifecycle, including managing retrieved data and previous interactions. Unlike prompt engineering, which emphasizes text formatting and instruction writing, context engineering dynamically aggregates data during each model request. This approach necessitates the design of automated systems for data collection and information filtering, enhancing effectiveness in production environments.

Agents

AI Agent Memory: Types, Storage, and How To Implement It

7 Jul 202622 Muharram 1448 AH

Agents

7 Jul 202622 Muharram 1448 AH

AI Agent Memory: Types, Storage, and How To Implement It

The concept of agent memory in artificial intelligence is explored, noting that each invocation begins without prior context. The guide covers types of memory, storage mechanisms, and implementation strategies within n8n. Despite expanding context windows, challenges in information retrieval persist, resulting in a decline in accuracy.

Agents

Ant Group's LingBot-Vision Claims 12 World Firsts: 1.1B Parameter Model Beats 7B DINOv3

7 Jul 202622 Muharram 1448 AH

Agents

7 Jul 202622 Muharram 1448 AH

Ant Group's LingBot-Vision Claims 12 World Firsts: 1.1B Parameter Model Beats 7B DINOv3

Ant Group has launched LingBot-Depth 2.0, achieving 12 world-first results across benchmark datasets. This model utilizes the LingBot-Vision foundation model, the first spatial-native vision model designed for embodied AI, released as open source. LingBot-Vision is trained for first-person spatial perception, embedding geometric understanding from the start. With approximately 1.1 billion parameters, it significantly outperforms DINOv3, which has 7 billion parameters, especially in depth estimation tasks.