🤖 AI Infrastructure - Venture Capital Funds

AI has crossed an important threshold in job readiness. We believe AI agents will soon be able to serve as an independent workforce across industries, likely within the next one or two years. We see this as a generational opportunity to make human work more valuable and human-centric.

Key Points:

Human-Centric Workforce: AI agents will soon be able to serve as an independent workforce across industries, making human work more valuable and human-centric.
Generational Opportunity: This shift presents a generational opportunity to redefine the role of humans in the workforce and create new value streams.
Near-Term Impact: We expect this impact to occur within the next one or two years, with significant implications for industries and economies worldwide.

🔗 Resources:

Original post ↗
Original source
NeoCognition
AI job readiness and human-centric workforce

🚀 AI Infrastructure - Marigold V2

Sweet! Marigold V2 is in ComfyUI now. It turns regular 2D images into ultra-detailed 3D depth maps, handling surface normals/material properties in a single step, and is super fast.

Key Points:

Ultra-Detailed 3D Depth Maps: Marigold V2 generates ultra-detailed 3D depth maps from regular 2D images, with a single step process.
Surface Normals/Material Properties: The model handles surface normals and material properties simultaneously, providing a more accurate representation of the scene.
Super Fast: Marigold V2 is designed to be fast, making it suitable for real-time applications.

🔗 Resources:

Original post ↗
Original source
ComfyUI
Marigold V2

🤖 AI Infrastructure - Latent Interface Training (LIT)

Are you training your VLAs or WAMs correctly? Your action expert may be learning vision–action shortcuts that undermine generalization beyond the training distribution. Introducing Latent Interface Training (LIT ): Learn to act first, then learn how to use vision.

Key Points:

Latent Interface Training (LIT): LIT is a new training approach that focuses on learning to act first, then learning how to use vision, to improve generalization.
Vision-Action Shortcuts: Traditional training methods may lead to vision-action shortcuts that undermine generalization beyond the training distribution.
Improved Generalization: LIT aims to improve generalization by decoupling action and vision learning.

🔗 Resources:

Original post ↗
Original source
DJiafei
Latent Interface Training (LIT)

🚀 AI Infrastructure - DeepSeek-V4.1-Flash

The DeepSeek-4.1-Flash technical report is now on @ChapterPal : https:// chapterpal.com/s/bd5db2ef/dee pseek-v41-flash-pushing-the-limits-of-kv-cache-compression … DeepSeek-V4.1-Flash addresses the rising cost of long-horizon agent workloads, where long inputs drive up prefill computation and large KV caches strain high-bandwidth memory and storage.

Key Points:

DeepSeek-V4.1-Flash: DeepSeek-V4.1-Flash is a new technical report that addresses the rising cost of long-horizon agent workloads.
KV Cache Compression: The report focuses on pushing the limits of KV cache compression to reduce memory and storage requirements.
Improved Efficiency: DeepSeek-V4.1-Flash aims to improve efficiency in long-horizon agent workloads by reducing prefill computation and KV cache strain.

🔗 Resources:

Original post ↗
Original source
ChapterPal
DeepSeek-V4.1-Flash

🚀 AI Infrastructure - Inference Landscape

The inference landscape is going to get a lot more hybrid in the near future. We found that accuracy per joule of local models has improved 18x in just 16 months: 5.9x from hardware, 3.0x from model gains. Great in-depth cover by @FT : https:// ft.trib.al/eVI82ZQ @Avanika15

Key Points:

Hybrid Inference Landscape: The inference landscape is expected to become more hybrid, with a mix of local and cloud-based models.
Accuracy per Joule: Accuracy per joule of local models has improved 18x in just 16 months, driven by hardware and model gains.
Improved Efficiency: The improvement in accuracy per joule is expected to lead to more efficient inference workflows.

🔗 Resources:

Original post ↗
Original source
FT
Inference landscape

🚀 AI Infrastructure - Robots with Generalist Capabilities

Robots with generalist capabilities require us to rethink safety beyond collision avoidance and alignment. Check out our new position paper, with a taxonomy of emerging risks and research directions: https:// arxiv.org/abs/2609.06326 v1 …

Key Points:

Generalist Capabilities: Robots with generalist capabilities require a reevaluation of safety beyond traditional collision avoidance and alignment.
Taxonomy of Emerging Risks: The position paper provides a taxonomy of emerging risks and research directions for robots with generalist capabilities.
Improved Safety: The paper aims to improve safety by identifying and addressing emerging risks.

🔗 Resources:

Original post ↗
Original source
arXiv
Robots with generalist capabilities

🚀 AI Infrastructure - Safety and Collision Avoidance

Glad you are safe. Even earlier prediction of hazards, even faster reaction time and overall significantly better safety and collision avoidance coming as part of the next big upgrade (v15).

Key Points:

Improved Safety: The next big upgrade (v15) is expected to bring improved safety features, including earlier prediction of hazards and faster reaction time.
Better Collision Avoidance: The upgrade is also expected to improve collision avoidance, making it safer for robots to operate.
Significant Improvement: The upgrade is expected to bring significant improvements in safety and collision avoidance.

🔗 Resources:

Original post ↗
Original source
aelluswamy
Safety and collision avoidance

🚀 AI Infrastructure - Screenshot of Tool

Screenshot of the tool I'm currently working on. It's a tool that completely fits what I wanted: it visualizes which topics are using which Transport (SHM, UDPv4, etc...) when using Fast DDS.

Key Points:

Visualization Tool: The tool is designed to visualize which topics are using which transport protocols (SHM, UDPv4, etc...) when using Fast DDS.
Customization: The tool is tailored to fit the specific needs of the user, providing a customized visualization experience.
Improved Understanding: The tool aims to improve understanding of the underlying transport protocols and their usage.

🔗 Resources:

Original post ↗
Original post
dandelion1124
Screenshot of tool

🚀 AI Infrastructure - Human Abstractions

For me there is an important difference here in concepts, I don't think "human abstractions" is the same thing as anthropomorphism, for example, one thing is to say that these agents can cooperate, can strategize, show preferences, etc, because these are often concepts developed

Key Points:

Human Abstractions: Human abstractions refer to the concepts and ideas that humans use to understand and interact with the world.
Anthropomorphism: Anthropomorphism is the attribution of human-like qualities or characteristics to non-human entities, such as agents or machines.
Cooperation and Strategy: Human abstractions can enable cooperation and strategy between agents, allowing them to work together and make decisions.

🔗 Resources:

Original post ↗
Original source
tarantulae
Human abstractions