Hello {{first_name | AI enthusiast}}

SoftBank pivots from telecom to Telco AI Cloud, City Detect locks in a $13M Series A to algorithmically spot urban blight, and researchers unveil memory-efficient training like POET-X while warning about AI “poverty porn” in charity imagery.

Scroll down to catch the signals that matter.

The POET-X technical report introduces a large language model training method that replaces dense projections with scalable orthogonal transformations to dramatically reduce activation memory. This enables deeper, wider models on fixed hardware, improves throughput, and can cut training costs for frontier and domain-specific models without sacrificing accuracy.

Expect foundation model builders to aggressively test POET-X-style architectures to push parameter counts without proportional GPU budget growth.

A new paper argues that censored large language models are ideal testbeds for “secret knowledge” elicitation, since sensitive capabilities are present but hidden behind safety layers. Researchers explore jailbreak-style techniques and systematic prompting strategies to recover suppressed information, offering a controlled way to study misuse risks and evaluate alignment defenses.

Governments and labs can use censored models as laboratory environments for red-teaming dangerous capabilities before deployment.

Tech Policy Press reports a Berlin court ruling that compels X to give researchers structured access to platform data under the EU Digital Services Act. The decision clarifies platforms’ transparency obligations, strengthens independent scrutiny of algorithmic harms, and sets an early legal benchmark for DSA enforcement across member states.

Platform governance teams must treat researcher data access as a legal duty, not a discretionary transparency gesture.

Research from the University of East Anglia, covered by Civil Society, warns that AI-generated “poverty porn” images in fundraising campaigns reinforce harmful stereotypes and erode donor trust. The study highlights how synthetic depictions of suffering, produced for emotional impact, can misrepresent communities and undermine ethical storytelling in the charity sector.

Fundraisers will need explicit AI image ethics policies to protect dignity, accuracy, and long-term credibility with supporters.

SoftBank Corp announced a strategic shift from pure connectivity to AI infrastructure with its Telco AI Cloud vision. The company aims to orchestrate distributed intelligence across edge sites and centralized clouds, enabling low-latency inference, network-aware optimization, and AI-native services for enterprises and developers on top of telecom assets.

Telcos worldwide may follow SoftBank, turning 5G and fiber footprints into competitive AI compute and orchestration platforms.

Our Sunday Visitor details a panel examining how AI data centers reshape local communities through land use, water demand, energy consumption, and tax incentives. Speakers raised ecological concerns, economic tradeoffs, and questions about human dignity when digital infrastructure drives resource extraction without corresponding social investment or transparent public deliberation.

Expect growing alliances between environmental advocates, faith communities, and residents challenging unexamined AI infrastructure expansion.

UC San Diego’s CW3E announced HyMeshAI, a deep learning-enabled system that automatically generates three-dimensional adaptive meshes for high-resolution atmospheric models. By focusing grid detail where storms intensify or moisture converges, it promises more accurate forecasts with lower computational cost, supporting extreme weather prediction and climate-related decision-making.

National weather agencies and climate modelers gain a powerful tool to target compute where it most improves hazardous event forecasts.

City Detect closed a $13 million Series A to expand its AI system that analyzes imagery and sensor data to identify urban blight, code violations, and safety risks. City governments can use the platform to prioritize inspections, allocate cleanup resources, and monitor neighborhood conditions in near real time.

Urban management is shifting toward continuous, AI-driven “eyes on the street,” raising both efficiency and civil liberties questions.

An article in the U.S. Army NCO Journal argues that military AI must remain constrained by human leadership and ethics (Invalid URL). It emphasizes noncommissioned officers’ responsibility to question automated recommendations, safeguard moral agency, and ensure AI tools support, rather than replace, professional judgment in complex, high-stakes operational contexts.

Defense AI programs will increasingly pair technical advancement with training that centers ethical decision-making and human accountability.

Questions, hot takes, or a trend you think we missed? Hit reply with your strongest signal or blindspot and help shape tomorrow’s brief.