Amazon Adds Agentic Fine-Tuning to SageMaker — Support for Llama, Qwen, DeepSeek, and Nova
Amazon Web Services (AWS) has added a new agentic fine-tuning capability to SageMaker AI. The feature lets AI agents assist developers as they customize language models, and it supports major open models including Llama, Qwen, DeepSeek, and Nova.
Traditional fine-tuning has required specialized expertise for hyperparameter tuning and dataset preparation, but with the agentic approach, developers can adapt models to their purposes with far less effort. AWS has already been strengthening its enterprise AI services, and this addition appears to be part of that effort.
While no specific benchmarks or pricing details have been announced yet, the broad support for major open models makes this a move worth watching for enterprises that value freedom of model choice in multi-cloud environments.
MIT Technology Review Publishes a 'Blueprint for Strengthening Democracy with AI'
MIT Technology Review has published an essay analyzing AI's impact on democracy across three layers: epistemic, agentic, and institutional. The authors, Andrew Sorota and Josh Hendler (of Eric Schmidt's office), argue that AI is driving fundamental changes at each stage of how citizens acquire information, take action, and engage in collective governance.
On the **epistemic layer**, they stress that as people increasingly turn to AI to decide 'what is true,' ensuring the truthfulness of model outputs is an urgent priority. They also cite research suggesting that AI-assisted fact-checking could earn trust from more angles than human-produced checks.
On the **agentic layer**, they warn of a future where personal AI agents conduct research, communications, and lobbying on users' behalf. If an agent merely reinforces a user's existing opinions and shields them from uncomfortable information, it is not serving the user's interests.
On the **institutional layer**, they note that several states and municipalities are already experimenting with large-scale democratic deliberation mediated by AI. They also cite a study (published in Science) showing how AI mediators can help citizens find common ground.
One line captures the essay's core: 'If you don't design for democratic outcomes, you're designing for something else.'
US State Department Issues Global Warning Over DeepSeek AI Theft
The US State Department has issued a warning to its embassies and consulates worldwide over suspected theft of AI-related intellectual property by China's DeepSeek. Reported by Reuters, the move is based on allegations that DeepSeek has improperly acquired US AI technology.
The story drew 3 points on Hacker News, where it was discussed as a significant development at the intersection of national security and the AI technology race. The move suggests that the US-China battle for AI supremacy is moving beyond the stage of trade restrictions and shifting into concrete law enforcement and diplomatic channels.
DeepSeek has consistently maintained that its models were developed independently, but the US government's stance appears to be hardening.
Microsoft Announces Plan to Double AI Infrastructure in Two Years
Microsoft has revealed plans to double its AI infrastructure over the next two years. Reported by The Next Platform, the announcement points to massive data center investment aimed at keeping pace with surging AI workloads.
Microsoft has already built out substantial AI infrastructure through its partnership with OpenAI, and the doubling plan is seen as a response to growing demand for its Copilot product line and Azure AI services. Specific investment figures and facility details have not yet been announced.
With Google, Amazon, Meta, and others likewise pouring huge sums into AI infrastructure, the move can be seen as part of an industry-wide race to expand capacity.
WIRED: AI Hiring Algorithm Suspected of Blocking Job Interviews
WIRED has published an investigation into a healthcare student who may have been denied interview opportunities by an AI hiring algorithm. Armed with a knowledge of Python, the student at the center of the story spent six months investigating whether it was indeed an algorithm that kept rejecting their applications.
AI-driven hiring screening has been widely adopted for its efficiency, but the black-box nature of algorithmic decision criteria is fueling frustration among applicants. In specialized fields like healthcare, it raises questions about how aptitude assessment itself should work.
No legal conclusion has been reached so far, but the case could accelerate calls for transparency and accountability in AI hiring.
---