AI News

Follow the latest AI and automation news

15 Apr 202627 Shawwal 1447 AH

Rethinking AI TCO: Why Cost per Token Is the Only Metric That Matters

Traditional data centers have evolved from mere data storage and retrieval to becoming AI production facilities, with AI output as the primary goal. This shift necessitates a reevaluation of the economic factors related to AI infrastructure, including total cost of ownership. Organizations continue to overly emphasize maximum chip specifications and computing costs, leading to a mismatch between inputs and outputs. Cost per token represents the total cost of producing each delivered token, typically expressed as cost per million tokens.

Models

14 Apr 202626 Shawwal 1447 AH

What I’ve been building: ATOM Report, post-training course, finishing my book, and ongoing research

The ATOM Report has been released, measuring the open language model ecosystem and providing updates on adoption metrics. It details the rise of GPT-OSS and the influence of Chinese players like Moonshot and Z.ai, introducing the Relative Adoption Metric (RAM) for evaluating recent models. Additionally, the RLHF Book is now ready for pre-order, aimed at beginners in post-training language models. Initiated on May 20, 2024, the book has been sent to production, focusing on essential insights for newcomers in the field.

Business

14 Apr 202626 Shawwal 1447 AH

Workflow Automation vs. Orchestration: Architectural Differences That Matter at Scale

Workflow automation focuses on individual tasks, while orchestration manages multiple tasks. Automation is effective for simple processes like sending notifications and updating records, but it faces limitations in monitoring and coordination, making it less suitable for complex operations. In contrast, orchestration acts as a central control unit that oversees automated tasks across various domains, providing better coordination among processes. It allows for more effective error handling, making it ideal for operations that require comprehensive monitoring and service coordination.

💡

Gemini Robotics-ER 1.6: Powering real-world robotics tasks through enhanced embodied reasoning

What impact could these enhancements have on the future of robotics?

Models

The inevitable need for an open model consortium

11 Apr 202623 Shawwal 1447 AH

Models

11 Apr 202623 Shawwal 1447 AH

The inevitable need for an open model consortium

Percy Liang, a professor at Stanford University, discussed the Marine project, which aims to create fully open models. There is a trend towards forming a consortium of companies to finance a core set of open models, although the timeline for this is unclear. Neutron, a subsidiary of Nvidia, represents an effort to fund this direction within a single wealthy company. Recent months have seen significant changes in open model labs, with prominent figures leaving companies like Quora and Ai2. These shifts are not surprising, as the cost of keeping pace with advancements in artificial intelligence is rising.

Agents

National Robotics Week — Latest Physical AI Research, Breakthroughs and Resources

10 Apr 202622 Shawwal 1447 AH

Agents

10 Apr 202622 Shawwal 1447 AH

National Robotics Week — Latest Physical AI Research, Breakthroughs and Resources

During National Robotics Week, NVIDIA showcased advancements in integrating AI into the physical world, significantly impacting industries like agriculture and manufacturing. Key announcements at GTC included the introduction of open models like NVIDIA Isaac GR00T, which enable robots to understand natural language and perform complex tasks using vision language action reasoning. Additionally, the open-source physics engine Newton 1.0 was released, providing a reliable foundation for accurate robot manipulation.

Governance

Claude Mythos and misguided open-weight fearmongering

9 Apr 202621 Shawwal 1447 AH

Governance

9 Apr 202621 Shawwal 1447 AH

Claude Mythos and misguided open-weight fearmongering

This week, the Claude Mythos model was announced, showcasing strong capabilities, particularly in cybersecurity. However, this has sparked a backlash against open-weight models, with concerns that digital infrastructure isn't ready for an open version, potentially enabling attacks from multiple parties. Historically, open models were deemed dangerous when OpenAI withheld GPT-2 weights in 2019 and released GPT-4 in 2023. The current narrative conflates various unknowns into a broad policy recommendation that could undermine cybersecurity readiness.

Business

New Future of Work: AI is driving rapid change, uneven benefits

9 Apr 202621 Shawwal 1447 AH

Business

9 Apr 202621 Shawwal 1447 AH

New Future of Work: AI is driving rapid change, uneven benefits

The New Future of Work 2025 report reveals how artificial intelligence is transforming the workplace. It notes that organizations treating AI as a collaborative partner are reaping the most significant benefits. Over the past five years, changes in work dynamics have been documented, but this year's shift appears to be more pronounced. The report emphasizes that AI not only accelerates workflows but also influences creativity, decision-making, collaboration, and learning processes. Research indicates that individuals are using AI in uneven ways, resulting in varied creative experiences.

Business

Ideas: Steering AI toward the work future we want

9 Apr 202621 Shawwal 1447 AH

Business

9 Apr 202621 Shawwal 1447 AH

Ideas: Steering AI toward the work future we want

Microsoft has released the New Future of Work Report 2025, discussing AI's impact on work. The report aims to understand how work is reshaped by pandemic changes and hybrid work practices. It emphasizes the importance of viewing AI as a collaborator rather than just a tool. The report highlights that the future of work is actively shaped by consumers, not predetermined. It stresses the need for proactive strategies in adopting AI, ensuring it enhances productivity and innovation. This shift requires a reevaluation of how AI is integrated into work environments.

Agents

Orchestration vs. Choreography: Which One to Choose – or Use Both?

9 Apr 202621 Shawwal 1447 AH

Agents

9 Apr 202621 Shawwal 1447 AH

Orchestration vs. Choreography: Which One to Choose – or Use Both?

The distinction between orchestration and choreography in microservices systems is significant. Orchestration relies on a central controller to manage each workflow step, providing comprehensive visibility and precise control. In contrast, choreography operates without a central controller, allowing services to communicate through events and function independently. Choosing between these two models is a strategic decision that impacts system scalability, correction, and operation in production environments.